Detection Criteria to EstimateBluetooth-based Travel Time
Srinivas S. Pulugurtha, Ph.D., P.E.Md. Shah Imran, M.S.
Venkata R. Duddu, Ph.D., E.I.
2015 Annual Transportation Research ForumAtlanta, GA
March 12, 2015
Introduction
• Increasing travel demand and congestion• MAP-21 and performance-based decision making• Performance measures and information is useful to both
the road users and practitioners• Reliability / consistency / dependability
Measures of ReliabilityIndex Equation Index Equation
NCHRP definition SD of travel time λSkew
AASHTO definition and TranSystems definition
Probability on-time performance
Buffer time (BT) Variability TT85-TT15
Buffer time index (BTI)Variability TT80-TT20
First worst travel times over a month Variability TT70-TT30
Second worst travel times over a month
Acceptable travel time variation index P(Tavg+ATTV)
Planning time (PT) Desired travel time reduction index P(Tave-DTTR)
Planning time index (PTI) Travel time index (TTI)
Travel time variability (TTV)
Frequency of congestion
Percent of days/periods that are congested
Introduction (Cont.)
• Key - need travel time information to assess reliability• Inaccuracy in travel time estimates may cause losing
the trust of users of the transportation system• Important to understand and validate the accuracy of the
travel times from available means of collecting data• How to collect / obtain travel time data? Floating test car method License plate matching technique GPS Bluetooth detector
Gaining popularity
…
Bluetooth Detectors & Travel Time
• No test car• Low cost per unit of data,
continuous data collection andno disruption to traffic
• Can be used to conduct O-Dstudies, pedestrian & transitdelay, etc.
Bluetooth Detectors & Travel Time (Cont.)
• Acceptable accuracy to estimate thetravel time under homogeneoustraffic conditions
• Spacing between Bluetoothdetectors?
• Errors: ~10 sec detection rate Data varies with type of device ,
antenna capability / placementheight, and functional class
Pedestrian / bicycle / transit /other
Bluetooth Detector & Detection of Vehicles
Literature Review
• Matching MAC addresses can be used to report travel time effectively (Wasson et al., 2008)
• Requires ~10 seconds for discovery of all Bluetooth devices within range, which can be a source of error to estimate travel times (Malinovskiy et al. 2010, Puckett and Vickich, 2010)
• Issue of MAC address groups that are produced by the Bluetooth detectors can be addressed by utilizing the time stamp for the first MAC address in a group
Literature Review (Cont.)
• Accuracy of the travel speeds on freeways generated from the collected MAC addresses increases with the increase of the distance between Bluetooth detectors and the decrease of vehicle speed (Haghani et al., 2010)
• Detection area should be large enough for the detection of nearly all vehicles with Bluetooth-enabled devices traveling at different speeds (Malinovskiy et al., 2010)
• Pedestrians and bicyclists with detectable devices and buses with multiple Bluetooth devices onboard are sources of outliers (Malinovskiy et al., 2010)
Objectives
• Multiple detections, signal delay, and non-uniform traffic flow can cause errors in Bluetooth travel time estimates
• Investigate the accuracy of travel time estimates from Bluetooth detectors
• Compare with manually collected travel times for arterial and freeway segments
• Evaluate detection criteria and how they could affect travel time estimates
Data Acquisition / Collection for Evaluation
• Study area: Charlotte, NC• Manual & Bluetooth data
collection 6 test corridors Data was collected during peak hours
on two consecutive weekdays for each corridor
Route Number Route Name Type # of Lanes AADT
Bus Availability Speed Limit (mph)Weekdays Weekends
11 North Tryon Major Arterial 3 25,000-30,000 Yes Yes 45
12 South Blvd Arterial 2 20,000-25,000 Yes Yes 40
14 Providence Road Arterial 2 30,000-40,000 Yes Yes 45
20 Sharon Road Local 2 14,000-20,000 Yes No 35
22 Graham Street Rd Arterial 2 14,000-20,000 Yes Yes 45
I-85 Interstate 85 Freeway 4 30,000-60,000 No No 65
Arterial Street Data Collection
• Installed in signal cabinet controllers
• Antenna height ~10 feet• Raw data from USB
flash drives• Acyclica data processing
and filtering technique
Sample Bluetooth Detection Based Travel TimeArterial Street
Freeway Data Collection
• Installed in traffic monitoring camera boxes
• Antenna height ~12-15 feet
• Raw data from USB flash drives
• Acyclica data processing and filtering technique
Sample Bluetooth Detection Based Travel TimeFreeway
% of Samples by % Difference in Travel Time% Diff. N Tryon St N Graham St Providence Rd Queens Rd South Blvd
Percent Difference in Travel Times between 6-10 AM0-10 15.71 12.50 23.21 27.14 16.25
10-20 17.14 3.57 25.00 22.86 21.2520-30 17.14 1.79 10.71 14.29 17.5030-40 11.43 1.79 19.64 11.43 16.2540-50 5.71 5.36 5.36 11.43 5.00>50 32.86 75.00 16.07 12.86 23.75
Percent Difference in Travel Times between 11-1 PM0-10 20.00 0.00 25.00 13.33 25.00
10-20 16.67 0.00 25.00 20.00 12.5020-30 6.67 0.00 16.67 13.33 15.6330-40 13.33 0.00 4.17 23.33 15.6340-50 16.67 0.00 4.17 20.00 12.50>50 26.67 100.00 25.00 10.00 18.75
Percent Difference in Travel Times between 3-6 PM0-10 23.33 1.79 18.75 25.00 16.18
10-20 16.67 7.14 14.58 25.00 17.6520-30 20.00 5.36 14.58 5.00 23.5330-40 13.33 5.36 12.50 17.50 17.6540-50 3.33 8.93 25.00 15.00 8.82>50 23.33 71.43 14.58 12.50 16.18
Bluetooth Detector Zone & Travel Time Based on Detection Criteria
• Signal strength travel time: time difference between the time-stamps measured at the closest proximity to the detectors
• First detection travel time: time difference between 1st location entry time-stamp and 2nd location entry time-stamp
• Last detection travel time: time difference between 1st location exit time-stamp and 2nd location exit time-stamp
• Minimum travel time: time difference between 1st location exit time-stamp and 2nd location entry time-stamp
• Maximum travel time: time difference between 1st location entry time-stamp and 2nd location exit time-stamp
Bluetooth Detection Zone (Cont.)
Freeway Segment
Statistical Analysis - Arterial Street
Travel TimeManual vs. Bluetooth
Signal Strength
First Detection
Last Detection
Minimum Maximum
t-value -4.20 5.31 4.06 2.11 10.79p-value <0.01 <0.01 <0.01 <0.01 <0.01
Correlation coefficient 0.88 0.88 0.87 0.87 0.87Mean Squared Error 1,286.15 1,394.43 1,402.15 1,130.55 2,811.71Mean Absolute Error 29.14 30.56 30.13 26.53 45.49
Minimum Absolute Error 0.01 0.05 0.38 0.20 1.93Maximum Absolute Error 123.75 125.20 118.88 102.24 141.84
Root Mean Squared Error
35.86 37.34 37.45 33.62 53.03
Mean Absolute Percentage Error
0.19 0.20 0.19 0.16 0.29
Manual vs. Bluetooth - Arterial Street Travel Times
Statistical Analysis - Freeway
Travel TimeManual vs. Bluetooth
Signal Strength
First Detection
Last Detection
Minimum Maximum
t-value 1.30 2.05 1.39 5.19 -2.44p-value 0.20 0.04 0.17 <0.01 0.02
Correlation coefficient 0.71 0.73 0.69 0.73 0.69Mean Squared Error 670.78 555.08 874.64 633.23 807.00Mean Absolute Error 19.99 18.81 21.56 19.37 20.71
Minimum Absolute Error 0.01 0.01 0.01 0.01 0.01Maximum Absolute Error 96.08 69.51 141.54 72.94 144.41
Root Mean Squared Error
25.90 23.56 29.57 25.16 28.41
Mean Absolute Percentage Error
0.23 0.22 0.24 0.23 0.23
Manual vs. Bluetooth - Freeway Travel Times
Conclusions
• Researched a group of combinations based on the detection points including device signal strength time-stamp difference first to first detection last to last detection first to last detection last to first detection
• Arterial segment generates more outliers than the freeway segment• Relatively high % difference than for freeways
Conclusions (Cont.)
• For arterial streets, minimum travel time followed by signal strength travel time provided more accurate results than other criteria
• For freeway, first detection travel time followed by minimum travel time provided more accurate results than the other criteria
• Criteria to capture travel time using Bluetooth detectors do vary by functional class / road characteristics
Acknowledgements
• This workshop presentation is prepared based on information collected for a research project funded by the United States Department of Transportation – Office of the Assistant Secretary for Research and Technology (USDOT/OST-R) under Cooperative Agreement Number RITARS-12-H-UNCC.
Disclaimer
• The views, opinions, findings, and conclusions reflected in this presentation are the responsibility of the authors only and do not represent the official policy or position of the USDOT/OST-R, or any State, or the University of North Carolina at Charlotte or other entity. The authors are responsible for the facts and the accuracy of the data presented herein. This presentation does not constitute a standard, specification, or regulation.