a framework for incident detection and notification in vehicular ad hoc networks
DESCRIPTION
PhD defense by Mahmoud AbuelelaOld Dominion UniversityDepartment of Computer ScienceMarch 25, 2011TRANSCRIPT
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
A Framework For Incident Detection AndNotification In Vehicular Ad Hoc Networks
M. AbuelelaAdvisor: Prof. S. Olariu
Department of Computer ScienceOld Dominion University
Norfolk, VA, USA
March 2011
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Outline
1 Introduction
2 System Infrastructure
3 Automatic Incident DetectionWarming upProbabilistic technique
4 Data DisseminationTraffic analysisData dissemination in undivided roadsData dissemination in divided roads
5 Securing Data DisseminationBelts Clustering
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Vehicular Ad Hoc Networks
In the US, Vehicular Ad Hoc Networks(VANETs) are using75 MHz of spectrum in the 5.850 to 5.925 GHz bandspecially allocated by the US Federal CommunicationsCommission(FCC)
VANETs have a number of specific characteristics that setthem apart from traditional Mobile Ad Hoc Networks(MANETs)
1 Movement pattern : nodes in MANETs are assumed tobe slow and can move everywhere.
2 Scale: VANETs involve thousands of fast-movingvehicles over tens of miles of roadways and streets
3 Connectivity: while MANETs may experience transientperiods of loss of connectivity, in VANETs, extendedperiods of disconnection are the norm rather than theexception
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Vehicular Ad Hoc Networks
In the US, Vehicular Ad Hoc Networks(VANETs) are using75 MHz of spectrum in the 5.850 to 5.925 GHz bandspecially allocated by the US Federal CommunicationsCommission(FCC)
VANETs have a number of specific characteristics that setthem apart from traditional Mobile Ad Hoc Networks(MANETs)
1 Movement pattern : nodes in MANETs are assumed tobe slow and can move everywhere.
2 Scale: VANETs involve thousands of fast-movingvehicles over tens of miles of roadways and streets
3 Connectivity: while MANETs may experience transientperiods of loss of connectivity, in VANETs, extendedperiods of disconnection are the norm rather than theexception
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Vehicular Ad Hoc Networks
In the US, Vehicular Ad Hoc Networks(VANETs) are using75 MHz of spectrum in the 5.850 to 5.925 GHz bandspecially allocated by the US Federal CommunicationsCommission(FCC)
VANETs have a number of specific characteristics that setthem apart from traditional Mobile Ad Hoc Networks(MANETs)
1 Movement pattern : nodes in MANETs are assumed tobe slow and can move everywhere.
2 Scale: VANETs involve thousands of fast-movingvehicles over tens of miles of roadways and streets
3 Connectivity: while MANETs may experience transientperiods of loss of connectivity, in VANETs, extendedperiods of disconnection are the norm rather than theexception
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Vehicular Ad Hoc Networks
In the US, Vehicular Ad Hoc Networks(VANETs) are using75 MHz of spectrum in the 5.850 to 5.925 GHz bandspecially allocated by the US Federal CommunicationsCommission(FCC)
VANETs have a number of specific characteristics that setthem apart from traditional Mobile Ad Hoc Networks(MANETs)
1 Movement pattern : nodes in MANETs are assumed tobe slow and can move everywhere.
2 Scale: VANETs involve thousands of fast-movingvehicles over tens of miles of roadways and streets
3 Connectivity: while MANETs may experience transientperiods of loss of connectivity, in VANETs, extendedperiods of disconnection are the norm rather than theexception
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Vehicular Ad Hoc Networks
In the US, Vehicular Ad Hoc Networks(VANETs) are using75 MHz of spectrum in the 5.850 to 5.925 GHz bandspecially allocated by the US Federal CommunicationsCommission(FCC)
VANETs have a number of specific characteristics that setthem apart from traditional Mobile Ad Hoc Networks(MANETs)
1 Movement pattern : nodes in MANETs are assumed tobe slow and can move everywhere.
2 Scale: VANETs involve thousands of fast-movingvehicles over tens of miles of roadways and streets
3 Connectivity: while MANETs may experience transientperiods of loss of connectivity, in VANETs, extendedperiods of disconnection are the norm rather than theexception
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Collision Warning Systems
Congested highways due to traffic incidents cost over $75billion a year in lost worker productivity, over 8.4 billiongallons of fuel and an average of 119 persons died each day.
A number of VANETs based systems and prototypes havebeen proposed to alert drivers about possible trafficincidents.
Most of these techniques give individual vehicle theresponsibility for inferring/notifying about the presence of anincident
This invites a host of serious and well-documented securityattacks.
Several routing and data dissemination protocols have beenproposed in literature where very few of them took lack ofconnectivity into consideration
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Collision Warning Systems
Congested highways due to traffic incidents cost over $75billion a year in lost worker productivity, over 8.4 billiongallons of fuel and an average of 119 persons died each day.
A number of VANETs based systems and prototypes havebeen proposed to alert drivers about possible trafficincidents.
Most of these techniques give individual vehicle theresponsibility for inferring/notifying about the presence of anincident
This invites a host of serious and well-documented securityattacks.
Several routing and data dissemination protocols have beenproposed in literature where very few of them took lack ofconnectivity into consideration
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Collision Warning Systems
Congested highways due to traffic incidents cost over $75billion a year in lost worker productivity, over 8.4 billiongallons of fuel and an average of 119 persons died each day.
A number of VANETs based systems and prototypes havebeen proposed to alert drivers about possible trafficincidents.
Most of these techniques give individual vehicle theresponsibility for inferring/notifying about the presence of anincident
This invites a host of serious and well-documented securityattacks.
Several routing and data dissemination protocols have beenproposed in literature where very few of them took lack ofconnectivity into consideration
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Collision Warning Systems
Congested highways due to traffic incidents cost over $75billion a year in lost worker productivity, over 8.4 billiongallons of fuel and an average of 119 persons died each day.
A number of VANETs based systems and prototypes havebeen proposed to alert drivers about possible trafficincidents.
Most of these techniques give individual vehicle theresponsibility for inferring/notifying about the presence of anincident
This invites a host of serious and well-documented securityattacks.
Several routing and data dissemination protocols have beenproposed in literature where very few of them took lack ofconnectivity into consideration
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Collision Warning Systems
Congested highways due to traffic incidents cost over $75billion a year in lost worker productivity, over 8.4 billiongallons of fuel and an average of 119 persons died each day.
A number of VANETs based systems and prototypes havebeen proposed to alert drivers about possible trafficincidents.
Most of these techniques give individual vehicle theresponsibility for inferring/notifying about the presence of anincident
This invites a host of serious and well-documented securityattacks.
Several routing and data dissemination protocols have beenproposed in literature where very few of them took lack ofconnectivity into consideration
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Automatic Incident Detection
Many incident detection algorithms exist from simple patternrecognition to AI techniques.Most developed techniques rely mainly on trafficmeasurements acquired at inductive loop detectors (ILDs) orvideo detection cameras installed at regular spacing alongthe freeways.All of these techniques assign cars a passive role in thedetection process.Also, It is fundamentally difficult to detect non-blockingaccidents and those that occur under light load, as thedeviation from normal traffic patterns may be negligible
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Automatic Incident Detection
Many incident detection algorithms exist from simple patternrecognition to AI techniques.Most developed techniques rely mainly on trafficmeasurements acquired at inductive loop detectors (ILDs) orvideo detection cameras installed at regular spacing alongthe freeways.All of these techniques assign cars a passive role in thedetection process.Also, It is fundamentally difficult to detect non-blockingaccidents and those that occur under light load, as thedeviation from normal traffic patterns may be negligible
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Automatic Incident Detection
Many incident detection algorithms exist from simple patternrecognition to AI techniques.Most developed techniques rely mainly on trafficmeasurements acquired at inductive loop detectors (ILDs) orvideo detection cameras installed at regular spacing alongthe freeways.All of these techniques assign cars a passive role in thedetection process.Also, It is fundamentally difficult to detect non-blockingaccidents and those that occur under light load, as thedeviation from normal traffic patterns may be negligible
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Automatic Incident Detection
Many incident detection algorithms exist from simple patternrecognition to AI techniques.Most developed techniques rely mainly on trafficmeasurements acquired at inductive loop detectors (ILDs) orvideo detection cameras installed at regular spacing alongthe freeways.All of these techniques assign cars a passive role in thedetection process.Also, It is fundamentally difficult to detect non-blockingaccidents and those that occur under light load, as thedeviation from normal traffic patterns may be negligible
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Automatic Incident Detection
False positive alarms have been unacceptably high foroperational purposes. For example, 2% false positives perILD means simply thousands false alarms daily from many ofthem which discredit the otherwise useful AID
Traffic signs are localized with the incidents locations.
Thus, driver may not have many choices to avoid the incident(usually there is a single "detour" to avoid the incident)
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Automatic Incident Detection
False positive alarms have been unacceptably high foroperational purposes. For example, 2% false positives perILD means simply thousands false alarms daily from many ofthem which discredit the otherwise useful AID
Traffic signs are localized with the incidents locations.
Thus, driver may not have many choices to avoid the incident(usually there is a single "detour" to avoid the incident)
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Automatic Incident Detection
False positive alarms have been unacceptably high foroperational purposes. For example, 2% false positives perILD means simply thousands false alarms daily from many ofthem which discredit the otherwise useful AID
Traffic signs are localized with the incidents locations.
Thus, driver may not have many choices to avoid the incident(usually there is a single "detour" to avoid the incident)
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Our Framework
1 System architecture: an infrastructure that collectsinformation from passing vehicles.
2 Incident detection engine: uses accumulated information todetect potential incidents and traffic delays
3 Notification dissemination: once an accident detected,approaching drivers need to be alerted about its existence.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Our Framework
1 System architecture: an infrastructure that collectsinformation from passing vehicles.
2 Incident detection engine: uses accumulated information todetect potential incidents and traffic delays
3 Notification dissemination: once an accident detected,approaching drivers need to be alerted about its existence.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Our Framework
1 System architecture: an infrastructure that collectsinformation from passing vehicles.
2 Incident detection engine: uses accumulated information todetect potential incidents and traffic delays
3 Notification dissemination: once an accident detected,approaching drivers need to be alerted about its existence.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Our Framework
1 System architecture: an infrastructure that collectsinformation from passing vehicles.
2 Incident detection engine: uses accumulated information todetect potential incidents and traffic delays
3 Notification dissemination: once an accident detected,approaching drivers need to be alerted about its existence.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Our Framework
1 System architecture: an infrastructure that collectsinformation from passing vehicles.
2 Incident detection engine: uses accumulated information todetect potential incidents and traffic delays
3 Notification dissemination: once an accident detected,approaching drivers need to be alerted about its existence.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
System Infrastructure: NOTICE
The first component of our framework is an architecture forthe notification of traffic incidents, NOTICE for short.
Using sensor belts embedded in the roadway, each messageis associated with physical vehicle
NOTICE moves the responsibility for making decisions abouttraffic-related information to the infrastructure rather thanleaving those decisions with the vehicles, which may haveincomplete or incorrect knowledge.
Existing automatic incident detection techniques areenhanced with cars and drivers inputs while preserving theprivacy of drivers.
Drivers are alerted early enough about traffic incidents totake the appropriate decisions.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
System Infrastructure: NOTICE
The first component of our framework is an architecture forthe notification of traffic incidents, NOTICE for short.
Using sensor belts embedded in the roadway, each messageis associated with physical vehicle
NOTICE moves the responsibility for making decisions abouttraffic-related information to the infrastructure rather thanleaving those decisions with the vehicles, which may haveincomplete or incorrect knowledge.
Existing automatic incident detection techniques areenhanced with cars and drivers inputs while preserving theprivacy of drivers.
Drivers are alerted early enough about traffic incidents totake the appropriate decisions.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
System Infrastructure: NOTICE
The first component of our framework is an architecture forthe notification of traffic incidents, NOTICE for short.
Using sensor belts embedded in the roadway, each messageis associated with physical vehicle
NOTICE moves the responsibility for making decisions abouttraffic-related information to the infrastructure rather thanleaving those decisions with the vehicles, which may haveincomplete or incorrect knowledge.
Existing automatic incident detection techniques areenhanced with cars and drivers inputs while preserving theprivacy of drivers.
Drivers are alerted early enough about traffic incidents totake the appropriate decisions.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
System Infrastructure: NOTICE
The first component of our framework is an architecture forthe notification of traffic incidents, NOTICE for short.
Using sensor belts embedded in the roadway, each messageis associated with physical vehicle
NOTICE moves the responsibility for making decisions abouttraffic-related information to the infrastructure rather thanleaving those decisions with the vehicles, which may haveincomplete or incorrect knowledge.
Existing automatic incident detection techniques areenhanced with cars and drivers inputs while preserving theprivacy of drivers.
Drivers are alerted early enough about traffic incidents totake the appropriate decisions.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
System Infrastructure: NOTICE
The first component of our framework is an architecture forthe notification of traffic incidents, NOTICE for short.
Using sensor belts embedded in the roadway, each messageis associated with physical vehicle
NOTICE moves the responsibility for making decisions abouttraffic-related information to the infrastructure rather thanleaving those decisions with the vehicles, which may haveincomplete or incorrect knowledge.
Existing automatic incident detection techniques areenhanced with cars and drivers inputs while preserving theprivacy of drivers.
Drivers are alerted early enough about traffic incidents totake the appropriate decisions.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts model
Every two consecutive belts A and B share a secretsymmetric key µ(A,B, t) known only between A and B.
Adjacent belts are connected using wires under the median
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Vehicles Model
Car’s EDR is a tamper-resistant device much like thewell-known black-boxes on commercial aircraft.All of the vehicle’s sub-assemblies, including the GPS unit,speedometer, gas tank reading, and sensors for outsidetemperature, feed their own readings into the EDRCars are equipped with automatic lane detections and feedsits output to its EDRGiven a time interval I , the EDR stores information such asthe highest and lowest speed during I, the position and timeof the strongest deceleration during I, as well as locationlane changes informationDrivers can provide input to the EDR, using a simple menu,either through a dashboard console or through verbal input
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Vehicles Model
Car’s EDR is a tamper-resistant device much like thewell-known black-boxes on commercial aircraft.All of the vehicle’s sub-assemblies, including the GPS unit,speedometer, gas tank reading, and sensors for outsidetemperature, feed their own readings into the EDRCars are equipped with automatic lane detections and feedsits output to its EDRGiven a time interval I , the EDR stores information such asthe highest and lowest speed during I, the position and timeof the strongest deceleration during I, as well as locationlane changes informationDrivers can provide input to the EDR, using a simple menu,either through a dashboard console or through verbal input
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Vehicles Model
Car’s EDR is a tamper-resistant device much like thewell-known black-boxes on commercial aircraft.All of the vehicle’s sub-assemblies, including the GPS unit,speedometer, gas tank reading, and sensors for outsidetemperature, feed their own readings into the EDRCars are equipped with automatic lane detections and feedsits output to its EDRGiven a time interval I , the EDR stores information such asthe highest and lowest speed during I, the position and timeof the strongest deceleration during I, as well as locationlane changes informationDrivers can provide input to the EDR, using a simple menu,either through a dashboard console or through verbal input
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Vehicles Model
Car’s EDR is a tamper-resistant device much like thewell-known black-boxes on commercial aircraft.All of the vehicle’s sub-assemblies, including the GPS unit,speedometer, gas tank reading, and sensors for outsidetemperature, feed their own readings into the EDRCars are equipped with automatic lane detections and feedsits output to its EDRGiven a time interval I , the EDR stores information such asthe highest and lowest speed during I, the position and timeof the strongest deceleration during I, as well as locationlane changes informationDrivers can provide input to the EDR, using a simple menu,either through a dashboard console or through verbal input
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Vehicles Model
Car’s EDR is a tamper-resistant device much like thewell-known black-boxes on commercial aircraft.All of the vehicle’s sub-assemblies, including the GPS unit,speedometer, gas tank reading, and sensors for outsidetemperature, feed their own readings into the EDRCars are equipped with automatic lane detections and feedsits output to its EDRGiven a time interval I , the EDR stores information such asthe highest and lowest speed during I, the position and timeof the strongest deceleration during I, as well as locationlane changes informationDrivers can provide input to the EDR, using a simple menu,either through a dashboard console or through verbal input
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belt To Car Communications
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belt To Car Communications
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belt To Car Communications
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belt To Car Communications
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belt To Car Communications
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belt To Car Communications
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Our Framework
1 System architecture: an infrastructure that collectsinformation from passing vehicles.
2 Incident detection engine: uses accumulated information todetect potential incidents and traffic delays
3 Notification dissemination: once an accident detected,approaching drivers need to be alerted about its existence.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Warming Up: Basic Idea
A belt is responsible for collecting and managing Informationabout lane changes from passing vehicles
Each belt maintains a table RoadImage[m][n] where m isnumber of lanes and n is the distance between twoconsecutive belts.
When a belt receives an EDR from a passing vehicles, it willincrement RoadImage[i][j] for all positions (lane = i , position= j) that the car has passed over.
RoadImage[i][j] = x , means that x cars have passed overthe location (lane = i; position = j) recently.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Warming Up: Basic Idea
A belt is responsible for collecting and managing Informationabout lane changes from passing vehicles
Each belt maintains a table RoadImage[m][n] where m isnumber of lanes and n is the distance between twoconsecutive belts.
When a belt receives an EDR from a passing vehicles, it willincrement RoadImage[i][j] for all positions (lane = i , position= j) that the car has passed over.
RoadImage[i][j] = x , means that x cars have passed overthe location (lane = i; position = j) recently.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Warming Up: Basic Idea
A belt is responsible for collecting and managing Informationabout lane changes from passing vehicles
Each belt maintains a table RoadImage[m][n] where m isnumber of lanes and n is the distance between twoconsecutive belts.
When a belt receives an EDR from a passing vehicles, it willincrement RoadImage[i][j] for all positions (lane = i , position= j) that the car has passed over.
RoadImage[i][j] = x , means that x cars have passed overthe location (lane = i; position = j) recently.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Warming Up: Basic Idea
A belt is responsible for collecting and managing Informationabout lane changes from passing vehicles
Each belt maintains a table RoadImage[m][n] where m isnumber of lanes and n is the distance between twoconsecutive belts.
When a belt receives an EDR from a passing vehicles, it willincrement RoadImage[i][j] for all positions (lane = i , position= j) that the car has passed over.
RoadImage[i][j] = x , means that x cars have passed overthe location (lane = i; position = j) recently.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Problems With The Warming Up Technique
1 The previous technique still produces many false positivealarms, specially if we are interested in short detection time.
2 It does not consider the relation between positions at whicha vehicle has changed lane and the location of the accident.Whenever a change lane being made, a belt assumed thathaving an incident at each location is equiprobable.
3 The value of the detection threshold is very important yetdifficult to determine for each road.
4 More importantly, it is very difficult to incorporate otherparameters, drivers inputs or declaration, in the detectionprocess.
5 Calibration complexity and lack of universality (ortransferability)
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Problems With The Warming Up Technique
1 The previous technique still produces many false positivealarms, specially if we are interested in short detection time.
2 It does not consider the relation between positions at whicha vehicle has changed lane and the location of the accident.Whenever a change lane being made, a belt assumed thathaving an incident at each location is equiprobable.
3 The value of the detection threshold is very important yetdifficult to determine for each road.
4 More importantly, it is very difficult to incorporate otherparameters, drivers inputs or declaration, in the detectionprocess.
5 Calibration complexity and lack of universality (ortransferability)
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Problems With The Warming Up Technique
1 The previous technique still produces many false positivealarms, specially if we are interested in short detection time.
2 It does not consider the relation between positions at whicha vehicle has changed lane and the location of the accident.Whenever a change lane being made, a belt assumed thathaving an incident at each location is equiprobable.
3 The value of the detection threshold is very important yetdifficult to determine for each road.
4 More importantly, it is very difficult to incorporate otherparameters, drivers inputs or declaration, in the detectionprocess.
5 Calibration complexity and lack of universality (ortransferability)
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Problems With The Warming Up Technique
1 The previous technique still produces many false positivealarms, specially if we are interested in short detection time.
2 It does not consider the relation between positions at whicha vehicle has changed lane and the location of the accident.Whenever a change lane being made, a belt assumed thathaving an incident at each location is equiprobable.
3 The value of the detection threshold is very important yetdifficult to determine for each road.
4 More importantly, it is very difficult to incorporate otherparameters, drivers inputs or declaration, in the detectionprocess.
5 Calibration complexity and lack of universality (ortransferability)
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Problems With The Warming Up Technique
1 The previous technique still produces many false positivealarms, specially if we are interested in short detection time.
2 It does not consider the relation between positions at whicha vehicle has changed lane and the location of the accident.Whenever a change lane being made, a belt assumed thathaving an incident at each location is equiprobable.
3 The value of the detection threshold is very important yetdifficult to determine for each road.
4 More importantly, it is very difficult to incorporate otherparameters, drivers inputs or declaration, in the detectionprocess.
5 Calibration complexity and lack of universality (ortransferability)
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Probabilistic Technique: Basic Idea
Let Pr[I] to be the a priori probability (or belief) of an incidentI at a given position on the road
When a number of evidences E �s, correlated in both timeand position, about an existing incident are collected
A belt computes the posteriori probability of an incident I atthe given location as
Bel(I) =Pr[I] · Pr[E |I]
Pr[E ]= α · Pr[I] · Pr[E |I]
where Pr[E |I] is the likelihood, E represents any evidencesuch as changing lanes or passing over a road anomaly andα is computed by the law of total probability as
1Pr[I] · Pr[E |I] + Pr[I] · Pr[E |I]
.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Probabilistic Technique: Basic Idea
Let Pr[I] to be the a priori probability (or belief) of an incidentI at a given position on the road
When a number of evidences E �s, correlated in both timeand position, about an existing incident are collected
A belt computes the posteriori probability of an incident I atthe given location as
Bel(I) =Pr[I] · Pr[E |I]
Pr[E ]= α · Pr[I] · Pr[E |I]
where Pr[E |I] is the likelihood, E represents any evidencesuch as changing lanes or passing over a road anomaly andα is computed by the law of total probability as
1Pr[I] · Pr[E |I] + Pr[I] · Pr[E |I]
.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Probabilistic Technique: Basic Idea
Let Pr[I] to be the a priori probability (or belief) of an incidentI at a given position on the road
When a number of evidences E �s, correlated in both timeand position, about an existing incident are collected
A belt computes the posteriori probability of an incident I atthe given location as
Bel(I) =Pr[I] · Pr[E |I]
Pr[E ]= α · Pr[I] · Pr[E |I]
where Pr[E |I] is the likelihood, E represents any evidencesuch as changing lanes or passing over a road anomaly andα is computed by the law of total probability as
1Pr[I] · Pr[E |I] + Pr[I] · Pr[E |I]
.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Probabilistic Technique: Basic Idea
We are interested here in two types of incidents
Blocking incidents : Incidents that blocks oneor more lanes and approaching vehicles haveto change lane to avoid them. Trafficaccidents are good examples for this categoryNon-blocking incidents : Those that driversmay pass over if (s)he did not notice them.Potholes are very good example for thatwhere a driver may pass over, decelerate ,take it between wheels or change lane toavoid.
Without loss of generality, we will consider a couple ofevidences in each case. (lane change and driver input) forblocking incidents. (lane change, declaration, driver inputand pothole sensing) for non blocking ones.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Probabilistic Technique: Basic Idea
We are interested here in two types of incidents
Blocking incidents : Incidents that blocks oneor more lanes and approaching vehicles haveto change lane to avoid them. Trafficaccidents are good examples for this category
Non-blocking incidents : Those that driversmay pass over if (s)he did not notice them.Potholes are very good example for thatwhere a driver may pass over, decelerate ,take it between wheels or change lane toavoid.
Without loss of generality, we will consider a couple ofevidences in each case. (lane change and driver input) forblocking incidents. (lane change, declaration, driver inputand pothole sensing) for non blocking ones.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Probabilistic Technique: Basic Idea
We are interested here in two types of incidents
Blocking incidents : Incidents that blocks oneor more lanes and approaching vehicles haveto change lane to avoid them. Trafficaccidents are good examples for this categoryNon-blocking incidents : Those that driversmay pass over if (s)he did not notice them.Potholes are very good example for thatwhere a driver may pass over, decelerate ,take it between wheels or change lane toavoid.
Without loss of generality, we will consider a couple ofevidences in each case. (lane change and driver input) forblocking incidents. (lane change, declaration, driver inputand pothole sensing) for non blocking ones.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Probabilistic Technique: Basic Idea
We are interested here in two types of incidents
Blocking incidents : Incidents that blocks oneor more lanes and approaching vehicles haveto change lane to avoid them. Trafficaccidents are good examples for this categoryNon-blocking incidents : Those that driversmay pass over if (s)he did not notice them.Potholes are very good example for thatwhere a driver may pass over, decelerate ,take it between wheels or change lane toavoid.
Without loss of generality, we will consider a couple ofevidences in each case. (lane change and driver input) forblocking incidents. (lane change, declaration, driver inputand pothole sensing) for non blocking ones.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Blocking incident detection
Assume that an accident has occurred atposition p, let X be the random variable thatkeeps track of the position at which vehicleschange lanes. We show that
fX (x) =1√2Π
e−(x−p)2
2
Similarly, Let Y be a random variable that keeps track of theposition at which a driver would provide input after anincident We show that
fY (y) = I(D) · 1√2π
e−(p−y)2
2
Where I(D) is an indicator function that returns 1 if the driverprovided an input about the incident and returns 0 otherwise.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Blocking incident detection
Assume that an accident has occurred atposition p, let X be the random variable thatkeeps track of the position at which vehicleschange lanes. We show that
fX (x) =1√2Π
e−(x−p)2
2
Similarly, Let Y be a random variable that keeps track of theposition at which a driver would provide input after anincident We show that
fY (y) = I(D) · 1√2π
e−(p−y)2
2
Where I(D) is an indicator function that returns 1 if the driverprovided an input about the incident and returns 0 otherwise.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Blocking incident detection: UpdatingProbabilities
Each belt maintains a table TempProb[m][n] that keeps theprobabilities of having a blocking incidents on differentpositions
All probabilities are initialized to a very small valuerepresenting the original probability of having an incident onthat road
Let us assume that a belt has received an EDR from apassing car x
For each position p = (lane = i , position = j) that a car x haspassed over, a belt sets
TempProb[i][j] = Initial probability
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Blocking incident detection: UpdatingProbabilities
Each belt maintains a table TempProb[m][n] that keeps theprobabilities of having a blocking incidents on differentpositions
All probabilities are initialized to a very small valuerepresenting the original probability of having an incident onthat road
Let us assume that a belt has received an EDR from apassing car x
For each position p = (lane = i , position = j) that a car x haspassed over, a belt sets
TempProb[i][j] = Initial probability
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Blocking incident detection: UpdatingProbabilities
Each belt maintains a table TempProb[m][n] that keeps theprobabilities of having a blocking incidents on differentpositions
All probabilities are initialized to a very small valuerepresenting the original probability of having an incident onthat road
Let us assume that a belt has received an EDR from apassing car x
For each position p = (lane = i , position = j) that a car x haspassed over, a belt sets
TempProb[i][j] = Initial probability
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Blocking incident detection: UpdatingProbabilities
Each belt maintains a table TempProb[m][n] that keeps theprobabilities of having a blocking incidents on differentpositions
All probabilities are initialized to a very small valuerepresenting the original probability of having an incident onthat road
Let us assume that a belt has received an EDR from apassing car x
For each position p = (lane = i , position = j) that a car x haspassed over, a belt sets
TempProb[i][j] = Initial probability
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Blocking incident detection: UpdatingProbabilities
For each position p = (lane = i , position = j) that a car hasavoided, the following is performed
The belt computes the posteriori probability of having ablocking incident at p that forced x ’s driver to change lanebefore p as
TempProb[i][j] = α · TempProb[i][j] · Py,j
where Py,j is the probability of changing lane at position ygiven that an accident occurred at position j and α can becomputed by the law of total probability as described before.
If x ’s EDR shows a driver input about position p, we alsoupdate the posteriori probability as
TempProb[i][j] = α · TempProb[i][j] · Pz,j
where Pz,j is the probability that the driver provided input atposition z given that an incident did exist at position j .
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Blocking incident detection: UpdatingProbabilities
For each position p = (lane = i , position = j) that a car hasavoided, the following is performed
The belt computes the posteriori probability of having ablocking incident at p that forced x ’s driver to change lanebefore p as
TempProb[i][j] = α · TempProb[i][j] · Py,j
where Py,j is the probability of changing lane at position ygiven that an accident occurred at position j and α can becomputed by the law of total probability as described before.
If x ’s EDR shows a driver input about position p, we alsoupdate the posteriori probability as
TempProb[i][j] = α · TempProb[i][j] · Pz,j
where Pz,j is the probability that the driver provided input atposition z given that an incident did exist at position j .
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Blocking incident detection: UpdatingProbabilities
For each position p = (lane = i , position = j) that a car hasavoided, the following is performed
The belt computes the posteriori probability of having ablocking incident at p that forced x ’s driver to change lanebefore p as
TempProb[i][j] = α · TempProb[i][j] · Py,j
where Py,j is the probability of changing lane at position ygiven that an accident occurred at position j and α can becomputed by the law of total probability as described before.
If x ’s EDR shows a driver input about position p, we alsoupdate the posteriori probability as
TempProb[i][j] = α · TempProb[i][j] · Pz,j
where Pz,j is the probability that the driver provided input atposition z given that an incident did exist at position j .
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
The main difference between detecting blocking andnon-blocking incidents is that a driver may pass over anon-blocking incident if he could not avoid it while he has tochange lane to avoid the blocking one.Thus, not every driver will be able to, or have to, change laneto avoid passing over a pothole for example.let Z be a random variable that keeps track of the position atwhich vehicles change lanes to avoid passing over a pothole.We can write
fZ (z) = I(D).1√2π
e−(z−p)2
2
Where I(D) is an indicator function returning 1 if the drivercould change lane to avoid the pothole and 0 otherwiseTo enhance the detection process, vehicles are assumed tobe equipped with a sensing device that can detect when theypass over a speed bump or a pothole.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
The main difference between detecting blocking andnon-blocking incidents is that a driver may pass over anon-blocking incident if he could not avoid it while he has tochange lane to avoid the blocking one.Thus, not every driver will be able to, or have to, change laneto avoid passing over a pothole for example.let Z be a random variable that keeps track of the position atwhich vehicles change lanes to avoid passing over a pothole.We can write
fZ (z) = I(D).1√2π
e−(z−p)2
2
Where I(D) is an indicator function returning 1 if the drivercould change lane to avoid the pothole and 0 otherwiseTo enhance the detection process, vehicles are assumed tobe equipped with a sensing device that can detect when theypass over a speed bump or a pothole.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
The main difference between detecting blocking andnon-blocking incidents is that a driver may pass over anon-blocking incident if he could not avoid it while he has tochange lane to avoid the blocking one.Thus, not every driver will be able to, or have to, change laneto avoid passing over a pothole for example.let Z be a random variable that keeps track of the position atwhich vehicles change lanes to avoid passing over a pothole.We can write
fZ (z) = I(D).1√2π
e−(z−p)2
2
Where I(D) is an indicator function returning 1 if the drivercould change lane to avoid the pothole and 0 otherwiseTo enhance the detection process, vehicles are assumed tobe equipped with a sensing device that can detect when theypass over a speed bump or a pothole.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
The main difference between detecting blocking andnon-blocking incidents is that a driver may pass over anon-blocking incident if he could not avoid it while he has tochange lane to avoid the blocking one.Thus, not every driver will be able to, or have to, change laneto avoid passing over a pothole for example.let Z be a random variable that keeps track of the position atwhich vehicles change lanes to avoid passing over a pothole.We can write
fZ (z) = I(D).1√2π
e−(z−p)2
2
Where I(D) is an indicator function returning 1 if the drivercould change lane to avoid the pothole and 0 otherwiseTo enhance the detection process, vehicles are assumed tobe equipped with a sensing device that can detect when theypass over a speed bump or a pothole.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
Similar to the TempProb table, a belt maintains another tablePermProb[m][n]
Moreover, each belt maintains a list ManMade that storesinformation about man-made road anomalies like rail-roadcrossings and speed bump positions. This information isvery helpful to automatically filter out these anomalies whenreported by cars.
For each position p(i , j) that car x has avoided, assumingthat x ’s driver has made the last lane change at positiony < j .
x ’s driver might have changed lanes because of a pothole atp, so, the belt computes the posteriori probability as
permprob[i][j] = α · permprob[i][j] · Py,j
Where Py,j is the probability of changing lane at position ygiven that a pothole does exist at position j .
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
Similar to the TempProb table, a belt maintains another tablePermProb[m][n]
Moreover, each belt maintains a list ManMade that storesinformation about man-made road anomalies like rail-roadcrossings and speed bump positions. This information isvery helpful to automatically filter out these anomalies whenreported by cars.
For each position p(i , j) that car x has avoided, assumingthat x ’s driver has made the last lane change at positiony < j .
x ’s driver might have changed lanes because of a pothole atp, so, the belt computes the posteriori probability as
permprob[i][j] = α · permprob[i][j] · Py,j
Where Py,j is the probability of changing lane at position ygiven that a pothole does exist at position j .
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
Similar to the TempProb table, a belt maintains another tablePermProb[m][n]
Moreover, each belt maintains a list ManMade that storesinformation about man-made road anomalies like rail-roadcrossings and speed bump positions. This information isvery helpful to automatically filter out these anomalies whenreported by cars.
For each position p(i , j) that car x has avoided, assumingthat x ’s driver has made the last lane change at positiony < j .
x ’s driver might have changed lanes because of a pothole atp, so, the belt computes the posteriori probability as
permprob[i][j] = α · permprob[i][j] · Py,j
Where Py,j is the probability of changing lane at position ygiven that a pothole does exist at position j .
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
Similar to the TempProb table, a belt maintains another tablePermProb[m][n]
Moreover, each belt maintains a list ManMade that storesinformation about man-made road anomalies like rail-roadcrossings and speed bump positions. This information isvery helpful to automatically filter out these anomalies whenreported by cars.
For each position p(i , j) that car x has avoided, assumingthat x ’s driver has made the last lane change at positiony < j .
x ’s driver might have changed lanes because of a pothole atp, so, the belt computes the posteriori probability as
permprob[i][j] = α · permprob[i][j] · Py,j
Where Py,j is the probability of changing lane at position ygiven that a pothole does exist at position j .
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
For each position p = (lane = i , position = j) that a car x haspassed over:
1 If x ’s EDR has a record for a suspected pothole atposition p, p /∈ ManMade , then the belt updatesposteriori probability as
PermProb[i][j] = α·PermProb[i][j]·Pr[Detection|Pothole]
2 If x ’s EDR shows that x has significantly deceleratedaround position p, p /∈ ManMade, this means that theremight be a pothole at position p that the driver wantedto avoid or reduce its effect on his car.
PermProb[i][j] = α·PermProb[i][j]·Pr[Decelerate|Pothole]
Where Pr[Reduce|Pothole] is the probability that thedriver slows down when he sees a pothole.
If x ’s EDR shows nothing about position p, two options exist.Either that position is clear or the driver could avoid thepothole.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
For each position p = (lane = i , position = j) that a car x haspassed over:
1 If x ’s EDR has a record for a suspected pothole atposition p, p /∈ ManMade , then the belt updatesposteriori probability as
PermProb[i][j] = α·PermProb[i][j]·Pr[Detection|Pothole]
2 If x ’s EDR shows that x has significantly deceleratedaround position p, p /∈ ManMade, this means that theremight be a pothole at position p that the driver wantedto avoid or reduce its effect on his car.
PermProb[i][j] = α·PermProb[i][j]·Pr[Decelerate|Pothole]
Where Pr[Reduce|Pothole] is the probability that thedriver slows down when he sees a pothole.
If x ’s EDR shows nothing about position p, two options exist.Either that position is clear or the driver could avoid thepothole.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
For each position p = (lane = i , position = j) that a car x haspassed over:
1 If x ’s EDR has a record for a suspected pothole atposition p, p /∈ ManMade , then the belt updatesposteriori probability as
PermProb[i][j] = α·PermProb[i][j]·Pr[Detection|Pothole]
2 If x ’s EDR shows that x has significantly deceleratedaround position p, p /∈ ManMade, this means that theremight be a pothole at position p that the driver wantedto avoid or reduce its effect on his car.
PermProb[i][j] = α·PermProb[i][j]·Pr[Decelerate|Pothole]
Where Pr[Reduce|Pothole] is the probability that thedriver slows down when he sees a pothole.
If x ’s EDR shows nothing about position p, two options exist.Either that position is clear or the driver could avoid thepothole.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
For each position p = (lane = i , position = j) that a car x haspassed over:
1 If x ’s EDR has a record for a suspected pothole atposition p, p /∈ ManMade , then the belt updatesposteriori probability as
PermProb[i][j] = α·PermProb[i][j]·Pr[Detection|Pothole]
2 If x ’s EDR shows that x has significantly deceleratedaround position p, p /∈ ManMade, this means that theremight be a pothole at position p that the driver wantedto avoid or reduce its effect on his car.
PermProb[i][j] = α·PermProb[i][j]·Pr[Decelerate|Pothole]
Where Pr[Reduce|Pothole] is the probability that thedriver slows down when he sees a pothole.
If x ’s EDR shows nothing about position p, two options exist.Either that position is clear or the driver could avoid thepothole.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
In contrast to our technique for accident detection, potholeprobabilities are never re-initialized after receiving an EDRfrom a passing vehicle.
However, if a belt continues updating pothole probabilitieswith the arrival of each EDR showing lane changes, falsepositive alarms will be eventually generated.
Therefore, it is important to re-initialize pothole probabilitiesafter some detection duration to avoid the detection of falsepotholes
If this detection duration is very short, shorter than the meandetection time, we may never detect any pothole.
On the other hand, If it is very long, false alarms would begenerated.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
In contrast to our technique for accident detection, potholeprobabilities are never re-initialized after receiving an EDRfrom a passing vehicle.
However, if a belt continues updating pothole probabilitieswith the arrival of each EDR showing lane changes, falsepositive alarms will be eventually generated.
Therefore, it is important to re-initialize pothole probabilitiesafter some detection duration to avoid the detection of falsepotholes
If this detection duration is very short, shorter than the meandetection time, we may never detect any pothole.
On the other hand, If it is very long, false alarms would begenerated.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
In contrast to our technique for accident detection, potholeprobabilities are never re-initialized after receiving an EDRfrom a passing vehicle.
However, if a belt continues updating pothole probabilitieswith the arrival of each EDR showing lane changes, falsepositive alarms will be eventually generated.
Therefore, it is important to re-initialize pothole probabilitiesafter some detection duration to avoid the detection of falsepotholes
If this detection duration is very short, shorter than the meandetection time, we may never detect any pothole.
On the other hand, If it is very long, false alarms would begenerated.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
In contrast to our technique for accident detection, potholeprobabilities are never re-initialized after receiving an EDRfrom a passing vehicle.
However, if a belt continues updating pothole probabilitieswith the arrival of each EDR showing lane changes, falsepositive alarms will be eventually generated.
Therefore, it is important to re-initialize pothole probabilitiesafter some detection duration to avoid the detection of falsepotholes
If this detection duration is very short, shorter than the meandetection time, we may never detect any pothole.
On the other hand, If it is very long, false alarms would begenerated.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Non Blocking Incidents detection: UpdatingProbabilities
In contrast to our technique for accident detection, potholeprobabilities are never re-initialized after receiving an EDRfrom a passing vehicle.
However, if a belt continues updating pothole probabilitieswith the arrival of each EDR showing lane changes, falsepositive alarms will be eventually generated.
Therefore, it is important to re-initialize pothole probabilitiesafter some detection duration to avoid the detection of falsepotholes
If this detection duration is very short, shorter than the meandetection time, we may never detect any pothole.
On the other hand, If it is very long, false alarms would begenerated.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Incident detection
While updating the probabilities using an EDR of a passingcar, a belt also checks for incidents
If TempProb[i][j] > T1 or PermProb[i][j] > T2 for any i and j ,an alarm is raised.
The larger the threshold is, the more conservative the belt is,the longer the time needed to detect incidents, the lessincident detection rate and fewer false alarms reported.
One of the strong points about our proposed techniques isthat the threshold is not a magic number that is very hard toset like most traditional AID techniques
Different threshold may be used for different sections/partsof the road
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Incident detection
While updating the probabilities using an EDR of a passingcar, a belt also checks for incidents
If TempProb[i][j] > T1 or PermProb[i][j] > T2 for any i and j ,an alarm is raised.
The larger the threshold is, the more conservative the belt is,the longer the time needed to detect incidents, the lessincident detection rate and fewer false alarms reported.
One of the strong points about our proposed techniques isthat the threshold is not a magic number that is very hard toset like most traditional AID techniques
Different threshold may be used for different sections/partsof the road
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Incident detection
While updating the probabilities using an EDR of a passingcar, a belt also checks for incidents
If TempProb[i][j] > T1 or PermProb[i][j] > T2 for any i and j ,an alarm is raised.
The larger the threshold is, the more conservative the belt is,the longer the time needed to detect incidents, the lessincident detection rate and fewer false alarms reported.
One of the strong points about our proposed techniques isthat the threshold is not a magic number that is very hard toset like most traditional AID techniques
Different threshold may be used for different sections/partsof the road
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Incident detection
While updating the probabilities using an EDR of a passingcar, a belt also checks for incidents
If TempProb[i][j] > T1 or PermProb[i][j] > T2 for any i and j ,an alarm is raised.
The larger the threshold is, the more conservative the belt is,the longer the time needed to detect incidents, the lessincident detection rate and fewer false alarms reported.
One of the strong points about our proposed techniques isthat the threshold is not a magic number that is very hard toset like most traditional AID techniques
Different threshold may be used for different sections/partsof the road
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Incident detection
While updating the probabilities using an EDR of a passingcar, a belt also checks for incidents
If TempProb[i][j] > T1 or PermProb[i][j] > T2 for any i and j ,an alarm is raised.
The larger the threshold is, the more conservative the belt is,the longer the time needed to detect incidents, the lessincident detection rate and fewer false alarms reported.
One of the strong points about our proposed techniques isthat the threshold is not a magic number that is very hard toset like most traditional AID techniques
Different threshold may be used for different sections/partsof the road
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Integration with existing techniques
For permanent incident detection, the proposed technique isnovel in detecting potholes and other permanent incidents
For temporal incident detection like accidents, the proposedBayesian-based technique would work well under non densetraffic.
However, when traffic becomes very dense and/or theincident blocks many lanes, vehicles will simply stuck behindthe incident and would not be able to continue and providetheir information to next belt .
Also, we believe that the market penetration will impact thedeployment of NOTICE
NOTICE can be integrated well with existing techniques fromliterature to provide improved performance as marketpenetration increases.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Integration with existing techniques
For permanent incident detection, the proposed technique isnovel in detecting potholes and other permanent incidents
For temporal incident detection like accidents, the proposedBayesian-based technique would work well under non densetraffic.
However, when traffic becomes very dense and/or theincident blocks many lanes, vehicles will simply stuck behindthe incident and would not be able to continue and providetheir information to next belt .
Also, we believe that the market penetration will impact thedeployment of NOTICE
NOTICE can be integrated well with existing techniques fromliterature to provide improved performance as marketpenetration increases.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Integration with existing techniques
For permanent incident detection, the proposed technique isnovel in detecting potholes and other permanent incidents
For temporal incident detection like accidents, the proposedBayesian-based technique would work well under non densetraffic.
However, when traffic becomes very dense and/or theincident blocks many lanes, vehicles will simply stuck behindthe incident and would not be able to continue and providetheir information to next belt .
Also, we believe that the market penetration will impact thedeployment of NOTICE
NOTICE can be integrated well with existing techniques fromliterature to provide improved performance as marketpenetration increases.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Integration with existing techniques
For permanent incident detection, the proposed technique isnovel in detecting potholes and other permanent incidents
For temporal incident detection like accidents, the proposedBayesian-based technique would work well under non densetraffic.
However, when traffic becomes very dense and/or theincident blocks many lanes, vehicles will simply stuck behindthe incident and would not be able to continue and providetheir information to next belt .
Also, we believe that the market penetration will impact thedeployment of NOTICE
NOTICE can be integrated well with existing techniques fromliterature to provide improved performance as marketpenetration increases.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Integration with existing techniques
For permanent incident detection, the proposed technique isnovel in detecting potholes and other permanent incidents
For temporal incident detection like accidents, the proposedBayesian-based technique would work well under non densetraffic.
However, when traffic becomes very dense and/or theincident blocks many lanes, vehicles will simply stuck behindthe incident and would not be able to continue and providetheir information to next belt .
Also, we believe that the market penetration will impact thedeployment of NOTICE
NOTICE can be integrated well with existing techniques fromliterature to provide improved performance as marketpenetration increases.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Performance evaluation: Performance indices
Detection rate:
DR =Number of incidents detectedTotal number of incident cases
X100%
False alarm rate
FPR =Number of false alarms
Total number of incident − free input patternsX100%
Mean detection time
MTTD =1n
�̇(td − to)
Where td and to are the detection and occurrence times ofan incident respectively.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Performance evaluation: Performance indices
Detection rate:
DR =Number of incidents detectedTotal number of incident cases
X100%
False alarm rate
FPR =Number of false alarms
Total number of incident − free input patternsX100%
Mean detection time
MTTD =1n
�̇(td − to)
Where td and to are the detection and occurrence times ofan incident respectively.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Performance evaluation: Performance indices
Detection rate:
DR =Number of incidents detectedTotal number of incident cases
X100%
False alarm rate
FPR =Number of false alarms
Total number of incident − free input patternsX100%
Mean detection time
MTTD =1n
�̇(td − to)
Where td and to are the detection and occurrence times ofan incident respectively.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Simulation Results: Blocking incidents
Under very sparse traffic, the probabilistic technique requires long time to collect sufficient number ofreports
Under dense traffic lane changes would be very difficult and cars would slow down or stop to makelane change
It is fundamentally difficult for existing techniques , including CA, to detect incidents under sparsetraffic as the deviation in traffic parameters is very small
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Simulation Results: Blocking incidents
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Simulation Results: Blocking incidents
For both CA and the deterministic technique. The probabilistic technique offers a zero false positivealarms, assuming 0.7 detection threshold, that makes it perfect and very trustable to traffic operators.
This is because the proposed technique re-initialize any position that a car has passed over. Thus,even if some evidences accumulated against some position, slow car for example, these evidenceswill be cancelled when a single car passes over these positions.
Under sparse to moderate traffic flows, CA could not detect any change in the traffic parameters andcould not even detect real incidents. So, it offers zero false positive rate.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Simulation Results: Blocking incidents
Drivers input provided a great enhancement to the detection process
As expected, as the detection threshold increases, more time would be needed by a belt to detect theexistence of an accident.
As also expected, more driver input means simply less time to detect the accident when it happens.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Simulation Results: Blocking incidents
It is not a surprise that detection rate decreases with the increase of detection threshold
However, even when the belt became very conservative, detection threshold of 0.9, it could stillprovide reasonable detection rate, for example more that 60% under driver input probability of only0.6.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Simulation Results: Blocking incidents
Down to about 70% penetration, the probabilistic technique has almost the same performance as100% case.
However, for smaller penetration, longer time would be needed to collect sufficient information fromEDR equipped vehicles.
On the other hand, the integrated technique required less detection time than the proposed techniquebecause it is enhanced with California based algorithm.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Simulation Results: Blocking incidents
The mean detection times increases slightly with the increase of distance between belts.
This is expected because cars have to travel longer distance to report their EDRs to next belt.
However, even when belts are installed with 5 km inter-spacing, detection time is still around 3minutes that simply means that NOTICE architecture is very efficient cost-wise overcoming one of thebasic limitations of ILDs.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Simulation Results: Non blocking incidents
For sparse traffic, longer time would be needed to accumulate more evidences until the probability ofhaving a pothole reaches the specified detection threshold.
However, as the traffic becomes more dense, more cars would be available to report about thepothole resulting in shorter mean detection time.
As an illustration, only about 40 seconds was needed to detect a pothole with detection threshold of0.9 which is very reasonable time for that conservative detection threshold.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Simulation Results: Non blocking incidents
As expected, the smaller the detection threshold the longer the pothole mean detection time as a beltneeds to accumulate enough evidences.
The mean detection time does not exceed 2 minutes even under large values for detection thresholds.Hence, we can safely choose a high detection threshold, > 0.9, to avoid generating false positivealarms.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Simulation Results: Non blocking incidents
As we already mentioned before, it is important to re-initialize pothole probabilities after somedetection duration time to avoid the detection of false potholes.
If this detection duration is very short, shorter than the mean detection time, we may never detect anypothole.
On the other hand, If it is very long, false alarms would be generated.
Zero false alarms are generated for detection duration up to 6 minutes. Joining this with the fact thatpothole detection time is usually less than a minute, we can choose detection duration time to bearound 5 minutes. Thus, we can use large detection thresholds and avoid generating false negativealarms.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Simulation Results: Non blocking incidents
Even for P(Detection/Pothole) around 0.5, the detection time would be less than a minute.
As P(Detection/Pothole) decreases, the mean detection time increases. However, we do not expectP(Detection/Pothole) to be less than 0.5, if not much higher.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Our Framework
1 System architecture: an infrastructure that collectsinformation from passing vehicles.
2 Incident detection engine: uses accumulated information todetect potential incidents and traffic delays
3 Notification dissemination: once an accident detected,approaching drivers need to be alerted about its existence.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Notification dissemination
Assume that belt D� is aware about the accident thatoccurred, based on cars reports and behaviorD� will inform D about this fact which in turn encrypt amessage with µ(D;C; t) and uploads it to car bThis messages will be routed to belt C, and belt C do thesame process again to belt BWhenever a belt receives a notification, it immediatelyinforms its neighbour in the other direction to alertapproaching carsFor generalization, we describe our disseminationtechniques between two moving cars.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Notification dissemination
Assume that belt D� is aware about the accident thatoccurred, based on cars reports and behaviorD� will inform D about this fact which in turn encrypt amessage with µ(D;C; t) and uploads it to car bThis messages will be routed to belt C, and belt C do thesame process again to belt BWhenever a belt receives a notification, it immediatelyinforms its neighbour in the other direction to alertapproaching carsFor generalization, we describe our disseminationtechniques between two moving cars.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Notification dissemination
Assume that belt D� is aware about the accident thatoccurred, based on cars reports and behaviorD� will inform D about this fact which in turn encrypt amessage with µ(D;C; t) and uploads it to car bThis messages will be routed to belt C, and belt C do thesame process again to belt BWhenever a belt receives a notification, it immediatelyinforms its neighbour in the other direction to alertapproaching carsFor generalization, we describe our disseminationtechniques between two moving cars.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Notification dissemination
Assume that belt D� is aware about the accident thatoccurred, based on cars reports and behaviorD� will inform D about this fact which in turn encrypt amessage with µ(D;C; t) and uploads it to car bThis messages will be routed to belt C, and belt C do thesame process again to belt BWhenever a belt receives a notification, it immediatelyinforms its neighbour in the other direction to alertapproaching carsFor generalization, we describe our disseminationtechniques between two moving cars.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Notification dissemination
Assume that belt D� is aware about the accident thatoccurred, based on cars reports and behaviorD� will inform D about this fact which in turn encrypt amessage with µ(D;C; t) and uploads it to car bThis messages will be routed to belt C, and belt C do thesame process again to belt BWhenever a belt receives a notification, it immediatelyinforms its neighbour in the other direction to alertapproaching carsFor generalization, we describe our disseminationtechniques between two moving cars.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
probability of large gaps in co-directional traffic
First, we provide an analytical proof that end to endconnectivity is not guaranteed on the road
We show that the probability of having a disconnection onthe road can be computed as
�m−1i=1 (−1)i+1�m−1
i
��m+n−i(d+1)−2m−2
��m+n−2
n
� .
Where m is number ofvehicles, (m + n) is the roadlength measured in averagevehicle size and d is themaximum effectivetransmission range.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
probability of large gaps in co-directional traffic
First, we provide an analytical proof that end to endconnectivity is not guaranteed on the road
We show that the probability of having a disconnection onthe road can be computed as
�m−1i=1 (−1)i+1�m−1
i
��m+n−i(d+1)−2m−2
��m+n−2
n
� .
Where m is number ofvehicles, (m + n) is the roadlength measured in averagevehicle size and d is themaximum effectivetransmission range.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
probability of large gaps in co-directional traffic
First, we provide an analytical proof that end to endconnectivity is not guaranteed on the road
We show that the probability of having a disconnection onthe road can be computed as
�m−1i=1 (−1)i+1�m−1
i
��m+n−i(d+1)−2m−2
��m+n−2
n
� .
Where m is number ofvehicles, (m + n) is the roadlength measured in averagevehicle size and d is themaximum effectivetransmission range.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Evaluating the expected size of a cluster
Thus, co-directional traffic is inherently partitioned intoclusters where cars within each cluster enjoy end-to-endconnectivity.
An important question is What is the expected size of acluster?
E[cluster_size] =m ·
�m+n−2n
��m+n−2
n
�+ (m − 1) ·
�m+n−d−3n
� .
Simulation resultsalso haveconfirmed ourfinding.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Evaluating the expected size of a cluster
Thus, co-directional traffic is inherently partitioned intoclusters where cars within each cluster enjoy end-to-endconnectivity.
An important question is What is the expected size of acluster?
E[cluster_size] =m ·
�m+n−2n
��m+n−2
n
�+ (m − 1) ·
�m+n−d−3n
� .
Simulation resultsalso haveconfirmed ourfinding.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Evaluating the expected size of a cluster
Thus, co-directional traffic is inherently partitioned intoclusters where cars within each cluster enjoy end-to-endconnectivity.
An important question is What is the expected size of acluster?
E[cluster_size] =m ·
�m+n−2n
��m+n−2
n
�+ (m − 1) ·
�m+n−d−3n
� .
Simulation resultsalso haveconfirmed ourfinding.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Evaluating the expected size of a cluster
Thus, co-directional traffic is inherently partitioned intoclusters where cars within each cluster enjoy end-to-endconnectivity.
An important question is What is the expected size of acluster?
E[cluster_size] =m ·
�m+n−2n
��m+n−2
n
�+ (m − 1) ·
�m+n−d−3n
� .
Simulation resultsalso haveconfirmed ourfinding.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Cluster stability
Since co-directional vehicles move at a small relative speedwith respect to each other, we expect clusters to be quitestable and easy to maintain
we have simulated a stretch of highway with two free trafficflow of 15 and 20 cars/km. In both cases, the differencebetween the highest and lowest speed is 15km/hour
This figure shows theaverage cluster sizes over15 minutes of simulationtime.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Cluster stability
Since co-directional vehicles move at a small relative speedwith respect to each other, we expect clusters to be quitestable and easy to maintain
we have simulated a stretch of highway with two free trafficflow of 15 and 20 cars/km. In both cases, the differencebetween the highest and lowest speed is 15km/hour
This figure shows theaverage cluster sizes over15 minutes of simulationtime.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Cluster stability
Since co-directional vehicles move at a small relative speedwith respect to each other, we expect clusters to be quitestable and easy to maintain
we have simulated a stretch of highway with two free trafficflow of 15 and 20 cars/km. In both cases, the differencebetween the highest and lowest speed is 15km/hour
This figure shows theaverage cluster sizes over15 minutes of simulationtime.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA - Motivation Example
Suppose that car a wishes todeliver some data to car g
Most of the existing protocolswaste the bandwidth bysending the packet from a to b.Then the packet will bepropagated through c, d andfinally e
Car e will keep the packet for a while until it meets car a towhich it will upload the packet Thus, the packet originated atcar a ends up at car a !!!
If car a knew from the beginning about this situation, it wasbetter to keep the packet instead of sending it to car b
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA - Motivation Example
Suppose that car a wishes todeliver some data to car g
Most of the existing protocolswaste the bandwidth bysending the packet from a to b.Then the packet will bepropagated through c, d andfinally e
Car e will keep the packet for a while until it meets car a towhich it will upload the packet Thus, the packet originated atcar a ends up at car a !!!
If car a knew from the beginning about this situation, it wasbetter to keep the packet instead of sending it to car b
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA - Motivation Example
Suppose that car a wishes todeliver some data to car g
Most of the existing protocolswaste the bandwidth bysending the packet from a to b.Then the packet will bepropagated through c, d andfinally e
Car e will keep the packet for a while until it meets car a towhich it will upload the packet Thus, the packet originated atcar a ends up at car a !!!
If car a knew from the beginning about this situation, it wasbetter to keep the packet instead of sending it to car b
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA - Motivation Example
Suppose that car a wishes todeliver some data to car g
Most of the existing protocolswaste the bandwidth bysending the packet from a to b.Then the packet will bepropagated through c, d andfinally e
Car e will keep the packet for a while until it meets car a towhich it will upload the packet Thus, the packet originated atcar a ends up at car a !!!
If car a knew from the beginning about this situation, it wasbetter to keep the packet instead of sending it to car b
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA - Overview
Because of the small relative speed of cars in a cluster, thecluster structure is assumed to remain stable and easy tomaintain.
This has motivated us to develop an Opportunistic PacketRelaying protocol (OPERA) for disconnected VANETs.
OPERA takes advantage of the relative stability ofco-directional clusters to maintain pro-actively routinginformation in each co-directional cluster as well asinformation about overlapping oncoming clusters.
Packet relaying is achieved, opportunistically, by acombination of data muling and local routing with the help ofoncoming clusters in a clever way to avoid possible delaysand bandwidth wasting.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA - Overview
Because of the small relative speed of cars in a cluster, thecluster structure is assumed to remain stable and easy tomaintain.
This has motivated us to develop an Opportunistic PacketRelaying protocol (OPERA) for disconnected VANETs.
OPERA takes advantage of the relative stability ofco-directional clusters to maintain pro-actively routinginformation in each co-directional cluster as well asinformation about overlapping oncoming clusters.
Packet relaying is achieved, opportunistically, by acombination of data muling and local routing with the help ofoncoming clusters in a clever way to avoid possible delaysand bandwidth wasting.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA - Overview
Because of the small relative speed of cars in a cluster, thecluster structure is assumed to remain stable and easy tomaintain.
This has motivated us to develop an Opportunistic PacketRelaying protocol (OPERA) for disconnected VANETs.
OPERA takes advantage of the relative stability ofco-directional clusters to maintain pro-actively routinginformation in each co-directional cluster as well asinformation about overlapping oncoming clusters.
Packet relaying is achieved, opportunistically, by acombination of data muling and local routing with the help ofoncoming clusters in a clever way to avoid possible delaysand bandwidth wasting.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA - Overview
Because of the small relative speed of cars in a cluster, thecluster structure is assumed to remain stable and easy tomaintain.
This has motivated us to develop an Opportunistic PacketRelaying protocol (OPERA) for disconnected VANETs.
OPERA takes advantage of the relative stability ofco-directional clusters to maintain pro-actively routinginformation in each co-directional cluster as well asinformation about overlapping oncoming clusters.
Packet relaying is achieved, opportunistically, by acombination of data muling and local routing with the help ofoncoming clusters in a clever way to avoid possible delaysand bandwidth wasting.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA - The vehicle and communicationmodel
Every 300 ms each vehicle sends a beacon, we call itCluster Maintenance Beacon (CMB), with a range of about200-300 m. This beacon contains information that allowsvehicles to handshake and synchronize
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA - The vehicle and communicationmodel
Every 300 ms each vehicle sends a beacon, we call itCluster Maintenance Beacon (CMB), with a range of about200-300 m. This beacon contains information that allowsvehicles to handshake and synchronize
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA - The vehicle and communicationmodel
Every 300 ms each vehicle sends a beacon, we call itCluster Maintenance Beacon (CMB), with a range of about200-300 m. This beacon contains information that allowsvehicles to handshake and synchronize
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA : Cluster Formation
Each car that has not received, within a certain time-outinterval, a beacon from a co-directional car in front of it,declares itself leader of the cluster and sends thisinformation in the next beacon to the cars behind it.
The message will be then multi-hopped, piggybacked inregular beacons, throughout the cluster
Every co-directional car that receives such a CMBunderstands that it belongs to the cluster named after theleader
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA : Cluster Formation
Each car that has not received, within a certain time-outinterval, a beacon from a co-directional car in front of it,declares itself leader of the cluster and sends thisinformation in the next beacon to the cars behind it.
The message will be then multi-hopped, piggybacked inregular beacons, throughout the cluster
Every co-directional car that receives such a CMBunderstands that it belongs to the cluster named after theleader
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA : Cluster Formation
Each car that has not received, within a certain time-outinterval, a beacon from a co-directional car in front of it,declares itself leader of the cluster and sends thisinformation in the next beacon to the cars behind it.
The message will be then multi-hopped, piggybacked inregular beacons, throughout the cluster
Every co-directional car that receives such a CMBunderstands that it belongs to the cluster named after theleader
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA : Cluster Formation
Each car that has not received, within a certain time-outinterval, a beacon from a co-directional car in front of it,declares itself leader of the cluster and sends thisinformation in the next beacon to the cars behind it.
The message will be then multi-hopped, piggybacked inregular beacons, throughout the cluster
Every co-directional car that receives such a CMBunderstands that it belongs to the cluster named after theleader
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA : Maintaining Cluster-RelatedInformation
A generic vehicle x in a cluster X maintains pro activelyinformation about the cluster to which it belongs as well asoverlapping clusters
Car x maintains1 Clusterid , 1st hop neighbours list2 A second record contains information about head and
tails of x’s cluster.3 A flag to determine weather there is any overlapping
cluster in advance or not and the expected time to loosesuch overlap
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA : Maintaining Cluster-RelatedInformation
A generic vehicle x in a cluster X maintains pro activelyinformation about the cluster to which it belongs as well asoverlapping clusters
Car x maintains1 Clusterid , 1st hop neighbours list2 A second record contains information about head and
tails of x’s cluster.3 A flag to determine weather there is any overlapping
cluster in advance or not and the expected time to loosesuch overlap
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA : Maintaining Cluster-RelatedInformation
A generic vehicle x in a cluster X maintains pro activelyinformation about the cluster to which it belongs as well asoverlapping clusters
Car x maintains1 Clusterid , 1st hop neighbours list2 A second record contains information about head and
tails of x’s cluster.3 A flag to determine weather there is any overlapping
cluster in advance or not and the expected time to loosesuch overlap
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA : Maintaining Cluster-RelatedInformation
A generic vehicle x in a cluster X maintains pro activelyinformation about the cluster to which it belongs as well asoverlapping clusters
Car x maintains1 Clusterid , 1st hop neighbours list2 A second record contains information about head and
tails of x’s cluster.3 A flag to determine weather there is any overlapping
cluster in advance or not and the expected time to loosesuch overlap
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA : Maintaining Cluster-RelatedInformation
A generic vehicle x in a cluster X maintains pro activelyinformation about the cluster to which it belongs as well asoverlapping clusters
Car x maintains1 Clusterid , 1st hop neighbours list2 A second record contains information about head and
tails of x’s cluster.3 A flag to determine weather there is any overlapping
cluster in advance or not and the expected time to loosesuch overlap
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: the Baseline Algorithm
Assumes that some vehicle a incluster A has a packet to relay to avehicle x in cluster X such that thefollowing conditions are satisfied:
1 a and x are in opposite lanes of traffic2 A
�X is a connected graph
The Baseline Algorithm, Baseline(a,A,x,X), is the workhorseof OPERA and works as follows .
If x is within range of a, the packet is delivered directly.
Otherwise, by consulting its routing table, list of all 1-hopneighbors, car a forwards the packet to a car (in either A orX ) that minimizes the hop-count to x and start the baselineagain
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: the Baseline Algorithm
Assumes that some vehicle a incluster A has a packet to relay to avehicle x in cluster X such that thefollowing conditions are satisfied:
1 a and x are in opposite lanes of traffic2 A
�X is a connected graph
The Baseline Algorithm, Baseline(a,A,x,X), is the workhorseof OPERA and works as follows .
If x is within range of a, the packet is delivered directly.
Otherwise, by consulting its routing table, list of all 1-hopneighbors, car a forwards the packet to a car (in either A orX ) that minimizes the hop-count to x and start the baselineagain
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: the Baseline Algorithm
Assumes that some vehicle a incluster A has a packet to relay to avehicle x in cluster X such that thefollowing conditions are satisfied:
1 a and x are in opposite lanes of traffic2 A
�X is a connected graph
The Baseline Algorithm, Baseline(a,A,x,X), is the workhorseof OPERA and works as follows .
If x is within range of a, the packet is delivered directly.
Otherwise, by consulting its routing table, list of all 1-hopneighbors, car a forwards the packet to a car (in either A orX ) that minimizes the hop-count to x and start the baselineagain
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: the Baseline Algorithm
Assumes that some vehicle a incluster A has a packet to relay to avehicle x in cluster X such that thefollowing conditions are satisfied:
1 a and x are in opposite lanes of traffic2 A
�X is a connected graph
The Baseline Algorithm, Baseline(a,A,x,X), is the workhorseof OPERA and works as follows .
If x is within range of a, the packet is delivered directly.
Otherwise, by consulting its routing table, list of all 1-hopneighbors, car a forwards the packet to a car (in either A orX ) that minimizes the hop-count to x and start the baselineagain
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: the Baseline Algorithm
Assumes that some vehicle a incluster A has a packet to relay to avehicle x in cluster X such that thefollowing conditions are satisfied:
1 a and x are in opposite lanes of traffic2 A
�X is a connected graph
The Baseline Algorithm, Baseline(a,A,x,X), is the workhorseof OPERA and works as follows .
If x is within range of a, the packet is delivered directly.
Otherwise, by consulting its routing table, list of all 1-hopneighbors, car a forwards the packet to a car (in either A orX ) that minimizes the hop-count to x and start the baselineagain
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: The General Algorithm(1/3)
General(a,A, x ,X ), assumes thatvehicle a in cluster A has a packetto relay to some oncoming vehicle xin cluster X but that the graph A
�X
is not connected.
Let B be the closest co-directional cluster to X that overlapswith A
Let D, if any, be the leftmost cluster among the clusters thatoverlaps with B
Let S be the graph induced by A, B and all the clustersoverlapping B. It is clear that S is a connected graph.
There are two cases to route the packet held by a to the lastcar in S, h(S).
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: The General Algorithm(1/3)
General(a,A, x ,X ), assumes thatvehicle a in cluster A has a packetto relay to some oncoming vehicle xin cluster X but that the graph A
�X
is not connected.
Let B be the closest co-directional cluster to X that overlapswith A
Let D, if any, be the leftmost cluster among the clusters thatoverlaps with B
Let S be the graph induced by A, B and all the clustersoverlapping B. It is clear that S is a connected graph.
There are two cases to route the packet held by a to the lastcar in S, h(S).
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: The General Algorithm(1/3)
General(a,A, x ,X ), assumes thatvehicle a in cluster A has a packetto relay to some oncoming vehicle xin cluster X but that the graph A
�X
is not connected.
Let B be the closest co-directional cluster to X that overlapswith A
Let D, if any, be the leftmost cluster among the clusters thatoverlaps with B
Let S be the graph induced by A, B and all the clustersoverlapping B. It is clear that S is a connected graph.
There are two cases to route the packet held by a to the lastcar in S, h(S).
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: The General Algorithm(1/3)
General(a,A, x ,X ), assumes thatvehicle a in cluster A has a packetto relay to some oncoming vehicle xin cluster X but that the graph A
�X
is not connected.
Let B be the closest co-directional cluster to X that overlapswith A
Let D, if any, be the leftmost cluster among the clusters thatoverlaps with B
Let S be the graph induced by A, B and all the clustersoverlapping B. It is clear that S is a connected graph.
There are two cases to route the packet held by a to the lastcar in S, h(S).
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: The General Algorithm(2 of 3)
Case 1: h(S) = h(A)
All that needs to be done is to routethe packet to the head of a’s cluster,h(A).
Case 2: h(S) �= h(A)The packet is routed using theBaseline Algorithm to one of cars,say, b that has connectivity to somecar in cluster D and then by theBaseline Algorithm again, to h(S).
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: The General Algorithm(2 of 3)
Case 1: h(S) = h(A)
All that needs to be done is to routethe packet to the head of a’s cluster,h(A).
Case 2: h(S) �= h(A)The packet is routed using theBaseline Algorithm to one of cars,say, b that has connectivity to somecar in cluster D and then by theBaseline Algorithm again, to h(S).
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: The General Algorithm(3/3)
Car a need not to be aware of all ofthese clusters. Car a formally doesthe following.
If it realizes that the oncoming cluster has some overlapingwith another co-directional cluster on the road,it can knowthat from beacons broadcasted by 1-hop cars in B, then thebaseline algorithm is applied until the packet reaches thatnext cluster.
Otherwise, it routes the packet to the header car of its owncluster.
The above algorithm is repeated at each node until thepacket reaches its destination.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: The General Algorithm(3/3)
Car a need not to be aware of all ofthese clusters. Car a formally doesthe following.
If it realizes that the oncoming cluster has some overlapingwith another co-directional cluster on the road,it can knowthat from beacons broadcasted by 1-hop cars in B, then thebaseline algorithm is applied until the packet reaches thatnext cluster.
Otherwise, it routes the packet to the header car of its owncluster.
The above algorithm is repeated at each node until thepacket reaches its destination.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
OPERA: The General Algorithm(3/3)
Car a need not to be aware of all ofthese clusters. Car a formally doesthe following.
If it realizes that the oncoming cluster has some overlapingwith another co-directional cluster on the road,it can knowthat from beacons broadcasted by 1-hop cars in B, then thebaseline algorithm is applied until the packet reaches thatnext cluster.
Otherwise, it routes the packet to the header car of its owncluster.
The above algorithm is repeated at each node until thepacket reaches its destination.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Cluster Maintenance Time
Joining this result with the fact that clusters are relatively stableand are not very large, we may claim that cars will have updated
information most of the time.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Percentage Of Wasted Bandwidth
Very little bandwidth is being wasted for most cases.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Message Overhead
DPP sends unnecessary messages when the co-directionalcluster has no overlap with two adjacent oncoming clusters.Moveover, in DPP, the message may be sent more thanonce.
OPERA avoids this overhead by sending a message only if itknows that it will achieve some progress
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Average Packet Delivery Time
A header car in DPP would send the packet to oncoming clusterand wait for confirmation from co-directional cluster on the road. Iffor any reason, this confirmation is lost or did not even exist, theheader car would wait then tries again.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Data dissemination in divided road - Motivationexample
We use same ideas we used in OPERA but in divided roadstaking speed into considerationSuppose that car s wishes to deliver some data to car eAgain, most of the existing protocols will waste thebandwidth by sending the packet from s to a. Then thepacket will be propagated through b, c and finally dCar d will keep the packet for a while until it meets car s towhich it will upload the packet Thus, the packet originated atcar s ends up at car s !!!If car s knew from the beginning about this situation, it wasbetter to keep the packet instead of sending it to car a
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Data dissemination in divided road - Motivationexample
We use same ideas we used in OPERA but in divided roadstaking speed into considerationSuppose that car s wishes to deliver some data to car eAgain, most of the existing protocols will waste thebandwidth by sending the packet from s to a. Then thepacket will be propagated through b, c and finally dCar d will keep the packet for a while until it meets car s towhich it will upload the packet Thus, the packet originated atcar s ends up at car s !!!If car s knew from the beginning about this situation, it wasbetter to keep the packet instead of sending it to car a
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Data dissemination in divided road - Motivationexample
We use same ideas we used in OPERA but in divided roadstaking speed into considerationSuppose that car s wishes to deliver some data to car eAgain, most of the existing protocols will waste thebandwidth by sending the packet from s to a. Then thepacket will be propagated through b, c and finally dCar d will keep the packet for a while until it meets car s towhich it will upload the packet Thus, the packet originated atcar s ends up at car s !!!If car s knew from the beginning about this situation, it wasbetter to keep the packet instead of sending it to car a
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Data dissemination in divided road - Motivationexample
We use same ideas we used in OPERA but in divided roadstaking speed into considerationSuppose that car s wishes to deliver some data to car eAgain, most of the existing protocols will waste thebandwidth by sending the packet from s to a. Then thepacket will be propagated through b, c and finally dCar d will keep the packet for a while until it meets car s towhich it will upload the packet Thus, the packet originated atcar s ends up at car s !!!If car s knew from the beginning about this situation, it wasbetter to keep the packet instead of sending it to car a
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Data dissemination in divided road - Motivationexample
We use same ideas we used in OPERA but in divided roadstaking speed into considerationSuppose that car s wishes to deliver some data to car eAgain, most of the existing protocols will waste thebandwidth by sending the packet from s to a. Then thepacket will be propagated through b, c and finally dCar d will keep the packet for a while until it meets car s towhich it will upload the packet Thus, the packet originated atcar s ends up at car s !!!If car s knew from the beginning about this situation, it wasbetter to keep the packet instead of sending it to car a
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
SODA Overview
Assume that car s carries a packet to be sent to a fixeddestination d .
If s has no connectivity to any car in front of it on the road, itwill simply carry the packet.
if connectivity exists, then s will have two options.1 If car s finds itself is expected to meet the destination
before all headers of its cluster, then it is better to keepthe packet in order to save communication resources.
2 if s found that the header of its cluster is expected tomeet the destination before s, then s will send thepacket to the furthest car using its list of single-hopneighbours
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
SODA Overview
Assume that car s carries a packet to be sent to a fixeddestination d .
If s has no connectivity to any car in front of it on the road, itwill simply carry the packet.
if connectivity exists, then s will have two options.1 If car s finds itself is expected to meet the destination
before all headers of its cluster, then it is better to keepthe packet in order to save communication resources.
2 if s found that the header of its cluster is expected tomeet the destination before s, then s will send thepacket to the furthest car using its list of single-hopneighbours
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
SODA Overview
Assume that car s carries a packet to be sent to a fixeddestination d .
If s has no connectivity to any car in front of it on the road, itwill simply carry the packet.
if connectivity exists, then s will have two options.1 If car s finds itself is expected to meet the destination
before all headers of its cluster, then it is better to keepthe packet in order to save communication resources.
2 if s found that the header of its cluster is expected tomeet the destination before s, then s will send thepacket to the furthest car using its list of single-hopneighbours
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
SODA Overview
Assume that car s carries a packet to be sent to a fixeddestination d .
If s has no connectivity to any car in front of it on the road, itwill simply carry the packet.
if connectivity exists, then s will have two options.1 If car s finds itself is expected to meet the destination
before all headers of its cluster, then it is better to keepthe packet in order to save communication resources.
2 if s found that the header of its cluster is expected tomeet the destination before s, then s will send thepacket to the furthest car using its list of single-hopneighbours
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
SODA Overview
Assume that car s carries a packet to be sent to a fixeddestination d .
If s has no connectivity to any car in front of it on the road, itwill simply carry the packet.
if connectivity exists, then s will have two options.1 If car s finds itself is expected to meet the destination
before all headers of its cluster, then it is better to keepthe packet in order to save communication resources.
2 if s found that the header of its cluster is expected tomeet the destination before s, then s will send thepacket to the furthest car using its list of single-hopneighbours
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Number of data messages
As expected, SODA sends fewer number of data messagesthan CAR for all cases especially under non dense trafficdensityIn a dense traffic ,in which connectivity exists most of thetime, SODA would send same number of data messages asCAR.However, total number of messages sent in CAR, control anddata, is twice those in this figure
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
CAR Wasted Bandwidth
CAR would waste much bandwidth time under non-densetraffic condition
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Average packet delivery time
SODA outperforms CAR under all traffic conditions andnumber of lanes.
This is because CAR waste much time in the discoveryprocess, sending control messages to the destination andthen back to the source
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
How far should a notification be propagated
A natural, yet important question is how far should anotification be propagated to alert drivers approachingincidents and what is the expected number of belts toreceive that update?
If the notification is propagated for a short distance, driversmay not be alerted early enough to make appropriatedecisions to avoid being blocked behind the accident.
On the other hand, propagating the notification for a longerdistance will waste a lot of expensive limited resources, likebandwidth, and moreover drivers far away from the accidentmay consider these alerts as false alarms
A reasonable answer for that question is that a belt shouldstop propagation of a message when it detects that cars onthe opposite direction are running normally and have notbeen backed up yet behind the accident.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
How far should a notification be propagated
A natural, yet important question is how far should anotification be propagated to alert drivers approachingincidents and what is the expected number of belts toreceive that update?
If the notification is propagated for a short distance, driversmay not be alerted early enough to make appropriatedecisions to avoid being blocked behind the accident.
On the other hand, propagating the notification for a longerdistance will waste a lot of expensive limited resources, likebandwidth, and moreover drivers far away from the accidentmay consider these alerts as false alarms
A reasonable answer for that question is that a belt shouldstop propagation of a message when it detects that cars onthe opposite direction are running normally and have notbeen backed up yet behind the accident.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
How far should a notification be propagated
A natural, yet important question is how far should anotification be propagated to alert drivers approachingincidents and what is the expected number of belts toreceive that update?
If the notification is propagated for a short distance, driversmay not be alerted early enough to make appropriatedecisions to avoid being blocked behind the accident.
On the other hand, propagating the notification for a longerdistance will waste a lot of expensive limited resources, likebandwidth, and moreover drivers far away from the accidentmay consider these alerts as false alarms
A reasonable answer for that question is that a belt shouldstop propagation of a message when it detects that cars onthe opposite direction are running normally and have notbeen backed up yet behind the accident.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
How far should a notification be propagated
A natural, yet important question is how far should anotification be propagated to alert drivers approachingincidents and what is the expected number of belts toreceive that update?
If the notification is propagated for a short distance, driversmay not be alerted early enough to make appropriatedecisions to avoid being blocked behind the accident.
On the other hand, propagating the notification for a longerdistance will waste a lot of expensive limited resources, likebandwidth, and moreover drivers far away from the accidentmay consider these alerts as false alarms
A reasonable answer for that question is that a belt shouldstop propagation of a message when it detects that cars onthe opposite direction are running normally and have notbeen backed up yet behind the accident.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
How far should a notification be propagated
Let Pi,i+1 be the probability of normal traffic between twoconsecutive belts i and i + 1By mapping the problem in hand to a typical COXdistribution, we show that:
1 The probability that the message propagates to exactly ibelts as
Ppropagation_exactly_i = Pi,i+1normal
i−1�
k=1(1 − Pk,k+1normal
)
2 The probability of the message being propagated for atleast i + 1 belts as
qat_least_i =i�
k=1(1 − Pk,k+1normal
)
3 The expected number of belts to receive the messagewould be
E =∞�
i=1(i ∗ Ppropagation_exactly_i )
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
How far should a notification be propagated
Let Pi,i+1 be the probability of normal traffic between twoconsecutive belts i and i + 1By mapping the problem in hand to a typical COXdistribution, we show that:
1 The probability that the message propagates to exactly ibelts as
Ppropagation_exactly_i = Pi,i+1normal
i−1�
k=1(1 − Pk,k+1normal
)
2 The probability of the message being propagated for atleast i + 1 belts as
qat_least_i =i�
k=1(1 − Pk,k+1normal
)
3 The expected number of belts to receive the messagewould be
E =∞�
i=1(i ∗ Ppropagation_exactly_i )
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
How far should a notification be propagated
Let Pi,i+1 be the probability of normal traffic between twoconsecutive belts i and i + 1By mapping the problem in hand to a typical COXdistribution, we show that:
1 The probability that the message propagates to exactly ibelts as
Ppropagation_exactly_i = Pi,i+1normal
i−1�
k=1(1 − Pk,k+1normal
)
2 The probability of the message being propagated for atleast i + 1 belts as
qat_least_i =i�
k=1(1 − Pk,k+1normal
)
3 The expected number of belts to receive the messagewould be
E =∞�
i=1(i ∗ Ppropagation_exactly_i )
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
How far should a notification be propagated
Let Pi,i+1 be the probability of normal traffic between twoconsecutive belts i and i + 1By mapping the problem in hand to a typical COXdistribution, we show that:
1 The probability that the message propagates to exactly ibelts as
Ppropagation_exactly_i = Pi,i+1normal
i−1�
k=1(1 − Pk,k+1normal
)
2 The probability of the message being propagated for atleast i + 1 belts as
qat_least_i =i�
k=1(1 − Pk,k+1normal
)
3 The expected number of belts to receive the messagewould be
E =∞�
i=1(i ∗ Ppropagation_exactly_i )
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
How far should a notification be propagated
Let Pi,i+1 be the probability of normal traffic between twoconsecutive belts i and i + 1By mapping the problem in hand to a typical COXdistribution, we show that:
1 The probability that the message propagates to exactly ibelts as
Ppropagation_exactly_i = Pi,i+1normal
i−1�
k=1(1 − Pk,k+1normal
)
2 The probability of the message being propagated for atleast i + 1 belts as
qat_least_i =i�
k=1(1 − Pk,k+1normal
)
3 The expected number of belts to receive the messagewould be
E =∞�
i=1(i ∗ Ppropagation_exactly_i )
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Our Framework
1 System architecture: an infrastructure that collectsinformation from passing vehicles.
2 Incident detection engine: uses accumulated information todetect potential incidents and traffic delays
3 Notification dissemination: once an accident detected,approaching drivers need to be alerted about its existence.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : Threat Model
messages exchanged between belts have to be secured againstdifferent kinds of attacks
Tracking a vehicle: knowing what sensors a car has mayreveal it’s driver identity
Replaying old messages: An attacker may record and replayold messages to passing cars
Eavesdropping: A malicious driver or person may try torecord all messages sent by belts or vehicles
Changing the message: An attacker may try to change orcorrupt a message.
Man in the middle: black hole in data dissemination.
Giving vehicle incorrect information: An attacker may throw asmall device over the road impersonating a belt and tellingcars incorrect information about fake incidents on the road
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : Threat Model
messages exchanged between belts have to be secured againstdifferent kinds of attacks
Tracking a vehicle: knowing what sensors a car has mayreveal it’s driver identity
Replaying old messages: An attacker may record and replayold messages to passing cars
Eavesdropping: A malicious driver or person may try torecord all messages sent by belts or vehicles
Changing the message: An attacker may try to change orcorrupt a message.
Man in the middle: black hole in data dissemination.
Giving vehicle incorrect information: An attacker may throw asmall device over the road impersonating a belt and tellingcars incorrect information about fake incidents on the road
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : Threat Model
messages exchanged between belts have to be secured againstdifferent kinds of attacks
Tracking a vehicle: knowing what sensors a car has mayreveal it’s driver identity
Replaying old messages: An attacker may record and replayold messages to passing cars
Eavesdropping: A malicious driver or person may try torecord all messages sent by belts or vehicles
Changing the message: An attacker may try to change orcorrupt a message.
Man in the middle: black hole in data dissemination.
Giving vehicle incorrect information: An attacker may throw asmall device over the road impersonating a belt and tellingcars incorrect information about fake incidents on the road
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : Threat Model
messages exchanged between belts have to be secured againstdifferent kinds of attacks
Tracking a vehicle: knowing what sensors a car has mayreveal it’s driver identity
Replaying old messages: An attacker may record and replayold messages to passing cars
Eavesdropping: A malicious driver or person may try torecord all messages sent by belts or vehicles
Changing the message: An attacker may try to change orcorrupt a message.
Man in the middle: black hole in data dissemination.
Giving vehicle incorrect information: An attacker may throw asmall device over the road impersonating a belt and tellingcars incorrect information about fake incidents on the road
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : Threat Model
messages exchanged between belts have to be secured againstdifferent kinds of attacks
Tracking a vehicle: knowing what sensors a car has mayreveal it’s driver identity
Replaying old messages: An attacker may record and replayold messages to passing cars
Eavesdropping: A malicious driver or person may try torecord all messages sent by belts or vehicles
Changing the message: An attacker may try to change orcorrupt a message.
Man in the middle: black hole in data dissemination.
Giving vehicle incorrect information: An attacker may throw asmall device over the road impersonating a belt and tellingcars incorrect information about fake incidents on the road
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : Threat Model
messages exchanged between belts have to be secured againstdifferent kinds of attacks
Tracking a vehicle: knowing what sensors a car has mayreveal it’s driver identity
Replaying old messages: An attacker may record and replayold messages to passing cars
Eavesdropping: A malicious driver or person may try torecord all messages sent by belts or vehicles
Changing the message: An attacker may try to change orcorrupt a message.
Man in the middle: black hole in data dissemination.
Giving vehicle incorrect information: An attacker may throw asmall device over the road impersonating a belt and tellingcars incorrect information about fake incidents on the road
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : Notations
{M}k is the result of encrypting message M with a key k .[C]k is the result of decrypting the cipher C using a key k .Publicx is the public key of node, vehicle or a belt, x .Privatex is the private key of node x .keyAB is a shared symmetric key that is known by both A andBM1|M2 is the resulting message of concatenating twomessages M1 and M2
A → B : M means that node A, vehicle or a belt, sends amessage M to node B. If B is *, this means that A isbroadcasting M.message: refers to an encrypted message that a belt wantsto communicate with next belt.notification: refers to a traffic-related information that a beltinforms vehicles about.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : Belt FunctionsAnd Assumptions
As has been suggested before, any message exchangedbetween two consecutive belts is encrypted with a sharedsecret symmetric key known between them.
A belt gives a message to more than one vehicle to increasethe probability of being received by next belt.
In addition to the shared symmetric key between any twoconsecutive belts, all belts within a given region share a timevariant private key, where the corresponding public key iswell known between cars. i.e. may be available over theinternet or broadcasted frequently over celulare network orsatellite communication using PKI certificates.
If a belt has a message to send to the next belt, it willtimestamp and sign that message using its private key.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : Belt FunctionsAnd Assumptions
As has been suggested before, any message exchangedbetween two consecutive belts is encrypted with a sharedsecret symmetric key known between them.
A belt gives a message to more than one vehicle to increasethe probability of being received by next belt.
In addition to the shared symmetric key between any twoconsecutive belts, all belts within a given region share a timevariant private key, where the corresponding public key iswell known between cars. i.e. may be available over theinternet or broadcasted frequently over celulare network orsatellite communication using PKI certificates.
If a belt has a message to send to the next belt, it willtimestamp and sign that message using its private key.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : Belt FunctionsAnd Assumptions
As has been suggested before, any message exchangedbetween two consecutive belts is encrypted with a sharedsecret symmetric key known between them.
A belt gives a message to more than one vehicle to increasethe probability of being received by next belt.
In addition to the shared symmetric key between any twoconsecutive belts, all belts within a given region share a timevariant private key, where the corresponding public key iswell known between cars. i.e. may be available over theinternet or broadcasted frequently over celulare network orsatellite communication using PKI certificates.
If a belt has a message to send to the next belt, it willtimestamp and sign that message using its private key.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : Belt FunctionsAnd Assumptions
As has been suggested before, any message exchangedbetween two consecutive belts is encrypted with a sharedsecret symmetric key known between them.
A belt gives a message to more than one vehicle to increasethe probability of being received by next belt.
In addition to the shared symmetric key between any twoconsecutive belts, all belts within a given region share a timevariant private key, where the corresponding public key iswell known between cars. i.e. may be available over theinternet or broadcasted frequently over celulare network orsatellite communication using PKI certificates.
If a belt has a message to send to the next belt, it willtimestamp and sign that message using its private key.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : VehicleFunctions And Assumptions
Vehicles are constantly getting belt’s public key for their areathrough satellite communication or cellular networks. Thesecan be done by the use of standard PKI certificates that iswidely used in distributing public keys in modern systems.
Vehicles accepts notifications about traffic condition frombelts or cars as long as they are signed using the belt’sprivate key and has not expired yet.
Each vehicle that receives a message directly from a belt willnever drop that message until it gives it to the next belt byitself. So, data mules still working behind the scene
While approaching a belt, not within the smallcommunication time, a vehicle encrypts its EDR data usingbelt’s public key before sending this data to next belt. Thus,only real belts can decrypt such information to do analysisand no attacker can understand EDR information
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : VehicleFunctions And Assumptions
Vehicles are constantly getting belt’s public key for their areathrough satellite communication or cellular networks. Thesecan be done by the use of standard PKI certificates that iswidely used in distributing public keys in modern systems.
Vehicles accepts notifications about traffic condition frombelts or cars as long as they are signed using the belt’sprivate key and has not expired yet.
Each vehicle that receives a message directly from a belt willnever drop that message until it gives it to the next belt byitself. So, data mules still working behind the scene
While approaching a belt, not within the smallcommunication time, a vehicle encrypts its EDR data usingbelt’s public key before sending this data to next belt. Thus,only real belts can decrypt such information to do analysisand no attacker can understand EDR information
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : VehicleFunctions And Assumptions
Vehicles are constantly getting belt’s public key for their areathrough satellite communication or cellular networks. Thesecan be done by the use of standard PKI certificates that iswidely used in distributing public keys in modern systems.
Vehicles accepts notifications about traffic condition frombelts or cars as long as they are signed using the belt’sprivate key and has not expired yet.
Each vehicle that receives a message directly from a belt willnever drop that message until it gives it to the next belt byitself. So, data mules still working behind the scene
While approaching a belt, not within the smallcommunication time, a vehicle encrypts its EDR data usingbelt’s public key before sending this data to next belt. Thus,only real belts can decrypt such information to do analysisand no attacker can understand EDR information
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination : VehicleFunctions And Assumptions
Vehicles are constantly getting belt’s public key for their areathrough satellite communication or cellular networks. Thesecan be done by the use of standard PKI certificates that iswidely used in distributing public keys in modern systems.
Vehicles accepts notifications about traffic condition frombelts or cars as long as they are signed using the belt’sprivate key and has not expired yet.
Each vehicle that receives a message directly from a belt willnever drop that message until it gives it to the next belt byitself. So, data mules still working behind the scene
While approaching a belt, not within the smallcommunication time, a vehicle encrypts its EDR data usingbelt’s public key before sending this data to next belt. Thus,only real belts can decrypt such information to do analysisand no attacker can understand EDR information
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
When a vehicle x passes over a belt A that has a message to besent to next belt B and/or a notification for x ,
A → x : Idx |msgid |{msg}KeyAB |notifications|expires_at |Asign
Idx : a unique identifier for each message sent by A. Thismay simply be an UUID stringmsgid : An UUID string that uniquely identifies the messageitself. so, msgid is unique for same message {msg}KeyAB
even if sent to multiple cars while Idx should be changing{msg}KeyAB : A message that needs to be communicatedsecurely with belt Bnotifications: Any notification that belt A would like x to beaware off, such as any incident/accident ahead on the roadexpires_at: A time after which this message should bediscarded (limit replaying messages attacks)Asign: A’s signature for that message.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
When a vehicle x passes over a belt A that has a message to besent to next belt B and/or a notification for x ,
A → x : Idx |msgid |{msg}KeyAB |notifications|expires_at |Asign
Idx : a unique identifier for each message sent by A. Thismay simply be an UUID stringmsgid : An UUID string that uniquely identifies the messageitself. so, msgid is unique for same message {msg}KeyAB
even if sent to multiple cars while Idx should be changing{msg}KeyAB : A message that needs to be communicatedsecurely with belt Bnotifications: Any notification that belt A would like x to beaware off, such as any incident/accident ahead on the roadexpires_at: A time after which this message should bediscarded (limit replaying messages attacks)Asign: A’s signature for that message.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
When a vehicle x passes over a belt A that has a message to besent to next belt B and/or a notification for x ,
A → x : Idx |msgid |{msg}KeyAB |notifications|expires_at |Asign
Idx : a unique identifier for each message sent by A. Thismay simply be an UUID stringmsgid : An UUID string that uniquely identifies the messageitself. so, msgid is unique for same message {msg}KeyAB
even if sent to multiple cars while Idx should be changing{msg}KeyAB : A message that needs to be communicatedsecurely with belt Bnotifications: Any notification that belt A would like x to beaware off, such as any incident/accident ahead on the roadexpires_at: A time after which this message should bediscarded (limit replaying messages attacks)Asign: A’s signature for that message.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
When a vehicle x passes over a belt A that has a message to besent to next belt B and/or a notification for x ,
A → x : Idx |msgid |{msg}KeyAB |notifications|expires_at |Asign
Idx : a unique identifier for each message sent by A. Thismay simply be an UUID stringmsgid : An UUID string that uniquely identifies the messageitself. so, msgid is unique for same message {msg}KeyAB
even if sent to multiple cars while Idx should be changing{msg}KeyAB : A message that needs to be communicatedsecurely with belt Bnotifications: Any notification that belt A would like x to beaware off, such as any incident/accident ahead on the roadexpires_at: A time after which this message should bediscarded (limit replaying messages attacks)Asign: A’s signature for that message.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
When a vehicle x passes over a belt A that has a message to besent to next belt B and/or a notification for x ,
A → x : Idx |msgid |{msg}KeyAB |notifications|expires_at |Asign
Idx : a unique identifier for each message sent by A. Thismay simply be an UUID stringmsgid : An UUID string that uniquely identifies the messageitself. so, msgid is unique for same message {msg}KeyAB
even if sent to multiple cars while Idx should be changing{msg}KeyAB : A message that needs to be communicatedsecurely with belt Bnotifications: Any notification that belt A would like x to beaware off, such as any incident/accident ahead on the roadexpires_at: A time after which this message should bediscarded (limit replaying messages attacks)Asign: A’s signature for that message.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
When a vehicle x passes over a belt A that has a message to besent to next belt B and/or a notification for x ,
A → x : Idx |msgid |{msg}KeyAB |notifications|expires_at |Asign
Idx : a unique identifier for each message sent by A. Thismay simply be an UUID stringmsgid : An UUID string that uniquely identifies the messageitself. so, msgid is unique for same message {msg}KeyAB
even if sent to multiple cars while Idx should be changing{msg}KeyAB : A message that needs to be communicatedsecurely with belt Bnotifications: Any notification that belt A would like x to beaware off, such as any incident/accident ahead on the roadexpires_at: A time after which this message should bediscarded (limit replaying messages attacks)Asign: A’s signature for that message.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
When a vehicle x passes over a belt A that has a message to besent to next belt B and/or a notification for x ,
A → x : Idx |msgid |{msg}KeyAB |notifications|expires_at |Asign
Idx : a unique identifier for each message sent by A. Thismay simply be an UUID stringmsgid : An UUID string that uniquely identifies the messageitself. so, msgid is unique for same message {msg}KeyAB
even if sent to multiple cars while Idx should be changing{msg}KeyAB : A message that needs to be communicatedsecurely with belt Bnotifications: Any notification that belt A would like x to beaware off, such as any incident/accident ahead on the roadexpires_at: A time after which this message should bediscarded (limit replaying messages attacks)Asign: A’s signature for that message.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
If x could successfully verify Asign for the message, thenotification would be displayed to x ’s driver console and xwill try to propagate the message to next cars
Let us assume for now that vehicle x has detected that thereare some other vehicles ahead of it on the road and vehicley has been selected by the dissemination technique to benext hop.
To avoid routing holes if y is a malicious attacker, x willauthenticate y first before dropping the message
x → y : Idx |msgid |{msg}KeyAB |notifications|expire_at |Asign
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
If x could successfully verify Asign for the message, thenotification would be displayed to x ’s driver console and xwill try to propagate the message to next cars
Let us assume for now that vehicle x has detected that thereare some other vehicles ahead of it on the road and vehicley has been selected by the dissemination technique to benext hop.
To avoid routing holes if y is a malicious attacker, x willauthenticate y first before dropping the message
x → y : Idx |msgid |{msg}KeyAB |notifications|expire_at |Asign
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
If x could successfully verify Asign for the message, thenotification would be displayed to x ’s driver console and xwill try to propagate the message to next cars
Let us assume for now that vehicle x has detected that thereare some other vehicles ahead of it on the road and vehicley has been selected by the dissemination technique to benext hop.
To avoid routing holes if y is a malicious attacker, x willauthenticate y first before dropping the message
x → y : Idx |msgid |{msg}KeyAB |notifications|expire_at |Asign
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
If x could successfully verify Asign for the message, thenotification would be displayed to x ’s driver console and xwill try to propagate the message to next cars
Let us assume for now that vehicle x has detected that thereare some other vehicles ahead of it on the road and vehicley has been selected by the dissemination technique to benext hop.
To avoid routing holes if y is a malicious attacker, x willauthenticate y first before dropping the message
x → y : Idx |msgid |{msg}KeyAB |notifications|expire_at |Asign
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
x is authenticating to y that it is a real vehicle that haspassed recently over the belt.
So, the following will be performed at y1 if A’s signature is valid and message not expired, go to
next step2 If this is a new message, using the message id, then
show the relevant notifications to y ’s console3 If y ’s position is beyond BPos, y has already passed the
belt, then it ignores that request. Otherwise continue tonext step.
4 y → x : Idy |{msg}KeyAB |notifications|expires_at |Asign
x can verify that y is a real car that has passed recently overA using Idy and A’s signature,.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
x is authenticating to y that it is a real vehicle that haspassed recently over the belt.
So, the following will be performed at y1 if A’s signature is valid and message not expired, go to
next step2 If this is a new message, using the message id, then
show the relevant notifications to y ’s console3 If y ’s position is beyond BPos, y has already passed the
belt, then it ignores that request. Otherwise continue tonext step.
4 y → x : Idy |{msg}KeyAB |notifications|expires_at |Asign
x can verify that y is a real car that has passed recently overA using Idy and A’s signature,.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
x is authenticating to y that it is a real vehicle that haspassed recently over the belt.
So, the following will be performed at y1 if A’s signature is valid and message not expired, go to
next step2 If this is a new message, using the message id, then
show the relevant notifications to y ’s console3 If y ’s position is beyond BPos, y has already passed the
belt, then it ignores that request. Otherwise continue tonext step.
4 y → x : Idy |{msg}KeyAB |notifications|expires_at |Asign
x can verify that y is a real car that has passed recently overA using Idy and A’s signature,.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
x is authenticating to y that it is a real vehicle that haspassed recently over the belt.
So, the following will be performed at y1 if A’s signature is valid and message not expired, go to
next step2 If this is a new message, using the message id, then
show the relevant notifications to y ’s console3 If y ’s position is beyond BPos, y has already passed the
belt, then it ignores that request. Otherwise continue tonext step.
4 y → x : Idy |{msg}KeyAB |notifications|expires_at |Asign
x can verify that y is a real car that has passed recently overA using Idy and A’s signature,.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
x is authenticating to y that it is a real vehicle that haspassed recently over the belt.
So, the following will be performed at y1 if A’s signature is valid and message not expired, go to
next step2 If this is a new message, using the message id, then
show the relevant notifications to y ’s console3 If y ’s position is beyond BPos, y has already passed the
belt, then it ignores that request. Otherwise continue tonext step.
4 y → x : Idy |{msg}KeyAB |notifications|expires_at |Asign
x can verify that y is a real car that has passed recently overA using Idy and A’s signature,.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
x is authenticating to y that it is a real vehicle that haspassed recently over the belt.
So, the following will be performed at y1 if A’s signature is valid and message not expired, go to
next step2 If this is a new message, using the message id, then
show the relevant notifications to y ’s console3 If y ’s position is beyond BPos, y has already passed the
belt, then it ignores that request. Otherwise continue tonext step.
4 y → x : Idy |{msg}KeyAB |notifications|expires_at |Asign
x can verify that y is a real car that has passed recently overA using Idy and A’s signature,.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
x is authenticating to y that it is a real vehicle that haspassed recently over the belt.
So, the following will be performed at y1 if A’s signature is valid and message not expired, go to
next step2 If this is a new message, using the message id, then
show the relevant notifications to y ’s console3 If y ’s position is beyond BPos, y has already passed the
belt, then it ignores that request. Otherwise continue tonext step.
4 y → x : Idy |{msg}KeyAB |notifications|expires_at |Asign
x can verify that y is a real car that has passed recently overA using Idy and A’s signature,.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
According to the previous protocol, it is very clear that thefollowing constraints are maintained and verified between x ,y and A.
1 x and y can verify that A is a real belt using A’ssignature and the expires_at values
2 y could verify that x is a car that has passed recentlyover A and the message to be propagated was reallysent by A
3 x , or any car in the dissemination process, could verifythat y is a real car that has passed recently over A.
One of the possible attacks is to have a malicious driver, D1who receives a real message from a belt and then stops bythe road replaying same message.
This is a limited attack as the message will expire.
Also, Belt B can easily identify duplicates using msgid
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
According to the previous protocol, it is very clear that thefollowing constraints are maintained and verified between x ,y and A.
1 x and y can verify that A is a real belt using A’ssignature and the expires_at values
2 y could verify that x is a car that has passed recentlyover A and the message to be propagated was reallysent by A
3 x , or any car in the dissemination process, could verifythat y is a real car that has passed recently over A.
One of the possible attacks is to have a malicious driver, D1who receives a real message from a belt and then stops bythe road replaying same message.
This is a limited attack as the message will expire.
Also, Belt B can easily identify duplicates using msgid
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
According to the previous protocol, it is very clear that thefollowing constraints are maintained and verified between x ,y and A.
1 x and y can verify that A is a real belt using A’ssignature and the expires_at values
2 y could verify that x is a car that has passed recentlyover A and the message to be propagated was reallysent by A
3 x , or any car in the dissemination process, could verifythat y is a real car that has passed recently over A.
One of the possible attacks is to have a malicious driver, D1who receives a real message from a belt and then stops bythe road replaying same message.
This is a limited attack as the message will expire.
Also, Belt B can easily identify duplicates using msgid
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
According to the previous protocol, it is very clear that thefollowing constraints are maintained and verified between x ,y and A.
1 x and y can verify that A is a real belt using A’ssignature and the expires_at values
2 y could verify that x is a car that has passed recentlyover A and the message to be propagated was reallysent by A
3 x , or any car in the dissemination process, could verifythat y is a real car that has passed recently over A.
One of the possible attacks is to have a malicious driver, D1who receives a real message from a belt and then stops bythe road replaying same message.
This is a limited attack as the message will expire.
Also, Belt B can easily identify duplicates using msgid
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
According to the previous protocol, it is very clear that thefollowing constraints are maintained and verified between x ,y and A.
1 x and y can verify that A is a real belt using A’ssignature and the expires_at values
2 y could verify that x is a car that has passed recentlyover A and the message to be propagated was reallysent by A
3 x , or any car in the dissemination process, could verifythat y is a real car that has passed recently over A.
One of the possible attacks is to have a malicious driver, D1who receives a real message from a belt and then stops bythe road replaying same message.
This is a limited attack as the message will expire.
Also, Belt B can easily identify duplicates using msgid
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
According to the previous protocol, it is very clear that thefollowing constraints are maintained and verified between x ,y and A.
1 x and y can verify that A is a real belt using A’ssignature and the expires_at values
2 y could verify that x is a car that has passed recentlyover A and the message to be propagated was reallysent by A
3 x , or any car in the dissemination process, could verifythat y is a real car that has passed recently over A.
One of the possible attacks is to have a malicious driver, D1who receives a real message from a belt and then stops bythe road replaying same message.
This is a limited attack as the message will expire.
Also, Belt B can easily identify duplicates using msgid
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination
According to the previous protocol, it is very clear that thefollowing constraints are maintained and verified between x ,y and A.
1 x and y can verify that A is a real belt using A’ssignature and the expires_at values
2 y could verify that x is a car that has passed recentlyover A and the message to be propagated was reallysent by A
3 x , or any car in the dissemination process, could verifythat y is a real car that has passed recently over A.
One of the possible attacks is to have a malicious driver, D1who receives a real message from a belt and then stops bythe road replaying same message.
This is a limited attack as the message will expire.
Also, Belt B can easily identify duplicates using msgid
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Securing Data Dissemination: Number ofcopies
Up to MD probability equals 60%, 5 copies would give a veryhigh probability that the message would reach next belt.
This figure shows that 9 copies of a message should besufficient even under misbehaved drivers probability of 80
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering
What happens if a belt B failed?
Unfortunately, if a belt B failed, then no other belt would beable to decrypt msgAB sent from A to B and hence, allincidents detected by A, or by any other belt before A, willnever be propagated beyond B. i.e. B becomes a black holein message propagation.
What if a belt is compromized?
The private key has to be changed to avoid having amalicious user sending incorrect data to drivers. Theprocess of changing that key will be very difficult if all beltsshare same private key.
The natural answer to the above two concerns is clustering
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering
What happens if a belt B failed?
Unfortunately, if a belt B failed, then no other belt would beable to decrypt msgAB sent from A to B and hence, allincidents detected by A, or by any other belt before A, willnever be propagated beyond B. i.e. B becomes a black holein message propagation.
What if a belt is compromized?
The private key has to be changed to avoid having amalicious user sending incorrect data to drivers. Theprocess of changing that key will be very difficult if all beltsshare same private key.
The natural answer to the above two concerns is clustering
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering
What happens if a belt B failed?
Unfortunately, if a belt B failed, then no other belt would beable to decrypt msgAB sent from A to B and hence, allincidents detected by A, or by any other belt before A, willnever be propagated beyond B. i.e. B becomes a black holein message propagation.
What if a belt is compromized?
The private key has to be changed to avoid having amalicious user sending incorrect data to drivers. Theprocess of changing that key will be very difficult if all beltsshare same private key.
The natural answer to the above two concerns is clustering
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering
What happens if a belt B failed?
Unfortunately, if a belt B failed, then no other belt would beable to decrypt msgAB sent from A to B and hence, allincidents detected by A, or by any other belt before A, willnever be propagated beyond B. i.e. B becomes a black holein message propagation.
What if a belt is compromized?
The private key has to be changed to avoid having amalicious user sending incorrect data to drivers. Theprocess of changing that key will be very difficult if all beltsshare same private key.
The natural answer to the above two concerns is clustering
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering
What happens if a belt B failed?
Unfortunately, if a belt B failed, then no other belt would beable to decrypt msgAB sent from A to B and hence, allincidents detected by A, or by any other belt before A, willnever be propagated beyond B. i.e. B becomes a black holein message propagation.
What if a belt is compromized?
The private key has to be changed to avoid having amalicious user sending incorrect data to drivers. Theprocess of changing that key will be very difficult if all beltsshare same private key.
The natural answer to the above two concerns is clustering
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering
Probability of having a black hole can be reduced.
Within a single cluster, belts share a secret symmetric key aswell as sharing same private key
Each belt in a cluster maintains a shared symmetric key withthe previous and next clusters on the road.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering
Probability of having a black hole can be reduced.
Within a single cluster, belts share a secret symmetric key aswell as sharing same private key
Each belt in a cluster maintains a shared symmetric key withthe previous and next clusters on the road.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering
Probability of having a black hole can be reduced.
Within a single cluster, belts share a secret symmetric key aswell as sharing same private key
Each belt in a cluster maintains a shared symmetric key withthe previous and next clusters on the road.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering
If a belt b2 in cluster B failed,any message sent by b1 canstill be decrypted at b3, sinceall belts within a cluster sharesame key
What if belt b3, which is supposed to encrypt the message,fails? Redundancy !
What if b2 is compromised ? the operator will have only tochange Bprivate, keyAB and keyBC which were known to b2
Clustering will also reduce the load of encryption/decryptionsince all belts within a cluster share same key
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering
If a belt b2 in cluster B failed,any message sent by b1 canstill be decrypted at b3, sinceall belts within a cluster sharesame key
What if belt b3, which is supposed to encrypt the message,fails? Redundancy !
What if b2 is compromised ? the operator will have only tochange Bprivate, keyAB and keyBC which were known to b2
Clustering will also reduce the load of encryption/decryptionsince all belts within a cluster share same key
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering
If a belt b2 in cluster B failed,any message sent by b1 canstill be decrypted at b3, sinceall belts within a cluster sharesame key
What if belt b3, which is supposed to encrypt the message,fails? Redundancy !
What if b2 is compromised ? the operator will have only tochange Bprivate, keyAB and keyBC which were known to b2
Clustering will also reduce the load of encryption/decryptionsince all belts within a cluster share same key
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering: cluster size
A natural yet important question is how large should acluster be?
On one extreme, if all belts form a single cluster then havingany belt compromised would require a tremendous workfrom the operator to manage installing new keys in all beltsand even worse, a malicious user would control all aspectsof NOTICE.
On the other extreme, if cluster size equals 1, no clustering,then a single belt failure means a black hole in messagepropagation between belts
We show that
Pblack_hole = 1 − (1 − Prf )
nm
Where pf is the probability that a single belt fails, n is total number of belts in a given section of the
highway and m is number of belts in a single cluster and r is number of replicas of the last belt
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering: cluster size
A natural yet important question is how large should acluster be?
On one extreme, if all belts form a single cluster then havingany belt compromised would require a tremendous workfrom the operator to manage installing new keys in all beltsand even worse, a malicious user would control all aspectsof NOTICE.
On the other extreme, if cluster size equals 1, no clustering,then a single belt failure means a black hole in messagepropagation between belts
We show that
Pblack_hole = 1 − (1 − Prf )
nm
Where pf is the probability that a single belt fails, n is total number of belts in a given section of the
highway and m is number of belts in a single cluster and r is number of replicas of the last belt
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering: cluster size
A natural yet important question is how large should acluster be?
On one extreme, if all belts form a single cluster then havingany belt compromised would require a tremendous workfrom the operator to manage installing new keys in all beltsand even worse, a malicious user would control all aspectsof NOTICE.
On the other extreme, if cluster size equals 1, no clustering,then a single belt failure means a black hole in messagepropagation between belts
We show that
Pblack_hole = 1 − (1 − Prf )
nm
Where pf is the probability that a single belt fails, n is total number of belts in a given section of the
highway and m is number of belts in a single cluster and r is number of replicas of the last belt
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering: cluster size
A natural yet important question is how large should acluster be?
On one extreme, if all belts form a single cluster then havingany belt compromised would require a tremendous workfrom the operator to manage installing new keys in all beltsand even worse, a malicious user would control all aspectsof NOTICE.
On the other extreme, if cluster size equals 1, no clustering,then a single belt failure means a black hole in messagepropagation between belts
We show that
Pblack_hole = 1 − (1 − Prf )
nm
Where pf is the probability that a single belt fails, n is total number of belts in a given section of the
highway and m is number of belts in a single cluster and r is number of replicas of the last belt
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Belts clustering: Blackhole probability
n m r Pf Pblack_hole1000 1 1 0.001 0.63230457521000 20 1 0.001 0.04879437181000 20 2 0.001 0.00004999881000 50 1 0.001 0.01981113521000 50 2 0.001 0.00001999981000 1 1 0.01 0.99995682881000 20 1 0.01 0.39499393291000 20 2 0.01 0.00498776961000 50 1 0.01 0.18209306241000 50 2 0.01 0.0019981011100 1 1 0.01 0.6339676587100 5 1 0.01 0.1820930624100 10 1 0.01 0.095617925050 1 1 0.01 0.394993932950 5 1 0.01 0.0956179250
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Conclusion
1 System architecture: an infrastructure that collectsinformation from passing vehicles.
2 Incident detection engine: uses accumulated information todetect potential incidents and traffic delays
3 Notification dissemination: once an accident detected,approaching drivers need to be alerted about its existence.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Future work and enhancements
Develop an AID technique to work even under very densetraffic by allowing stuck cars to send their EDRs to the beltby disseminating it.
Integrate NOTICE with the internet so that drivers can haveaccess to all incident information and travel times from homefor example.
With the governments strugglingly to install traffic lights andpatch the roads, It is a good idea to have NOTICE fundingitself through sending Adds and promotions for Gas stations,hotels, ... along with traffic information.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Future work and enhancements
Develop an AID technique to work even under very densetraffic by allowing stuck cars to send their EDRs to the beltby disseminating it.
Integrate NOTICE with the internet so that drivers can haveaccess to all incident information and travel times from homefor example.
With the governments strugglingly to install traffic lights andpatch the roads, It is a good idea to have NOTICE fundingitself through sending Adds and promotions for Gas stations,hotels, ... along with traffic information.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
Future work and enhancements
Develop an AID technique to work even under very densetraffic by allowing stuck cars to send their EDRs to the beltby disseminating it.
Integrate NOTICE with the internet so that drivers can haveaccess to all incident information and travel times from homefor example.
With the governments strugglingly to install traffic lights andpatch the roads, It is a good idea to have NOTICE fundingitself through sending Adds and promotions for Gas stations,hotels, ... along with traffic information.
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering
A FrameworkFor Incident
Detection AndNotification InVehicular Ad
Hoc Networks
M. AbuelelaAdvisor: Prof.
S. Olariu
Introduction
Infrastructure
IncidentDetectionWarming up
Probabilistictechnique
DataDisseminationTraffic analysis
Data disseminationin undivided roads
Data disseminationin divided roads
Securing DataDisseminationBelts Clustering