el nah raw y 04 are a based presentation
TRANSCRIPT
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
1/27
Using Area-based Presentations andMetrics for Localization Systems inWireless LANs
Eiman Elnahrawy, Xiaoyan Li, and Richard MartinDept. of Computer Science, Rutgers University
WLN, November 16th 2004
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
2/27
Dept. of Computer Science, Rutgers University
WLAN-Based Localization
Localization in indoor environments using802.11 and Fingerprinting
Numerous useful applications
Dual use infrastructure: a huge advantage
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
3/27
Dept. of Computer Science, Rutgers University
Background: Fingerprinting Localization
Classifiers/matching/learningapproaches
Offline phase:Collect training data (fingerprints)
Fingerprint vectors: [(x,y),SS]
Online phase:Match RSS to existingfingerprints probabilistically orusing a distance metric
[-80,-67,-50]
RSS
(x?,y?)
[(x,y),s1
,s2,s3]
[(x,y),s1,s2,s3]
[(x,y),s1,s2,s3]
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
4/27
Dept. of Computer Science, Rutgers University
Output:
A single location: the closest/best match
We call such approaches Point-based Localization
Examples:RADAR
Probabilistic approaches[Bahl00, Ladd02, Roos02, Smailagic02, Youssef03, Krishnan04]
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
5/27
Dept. of Computer Science, Rutgers University
Contributions: Area-based Localization
Returned answer is area/volumelikely to contain the localizedobject
Area is described by a set of tiles
Ability to describe uncertainty
Set of highly possible locations
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
6/27
Dept. of Computer Science, Rutgers University
Contributions: Area-based Localization
Show that it has critical advantages over point-basedlocalization
Introduce new performance metrics
Present two novel algorithms: SPM and ABP-c
Evaluate our algorithms and compare them againsttraditional point-based approaches
Related Work: different technologies/algorithms[Want92, Priyantha00, Doherty01, Niculescue01, Savvides01,Shang03, He03, Hazas03, Lorincz04]
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
7/27
Dept. of Computer Science, Rutgers University
Why Area-based?
Noise and systematic errors introduce position uncertainty
Areas improve systems ability to givemeaningful alternatives
A tool for understanding the confidence
Ability to trade Precision (area size) for Accuracy(distance the localized object is from the area)
Direct users in their search
Yields higher overall accuracy
Previous approaches that attempted to use areas only use themas intermediate result output still a single location
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
8/27
Dept. of Computer Science, Rutgers University
Area-based vs. Single-Location
Object can be in a single room or multiple rooms
Point-based to areasEnclosing circles --much larger
Rectangle? no longer point-based!
0200
0
10
20
30
40
50
60
70
0 200
80
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
9/27
Dept. of Computer Science, Rutgers University
Outline
Introduction, Motivations, and Related Work
Area-based vs. Point-based localization
Metrics
Localization AlgorithmsSimple Point Matching (SPM)
Area-based Probability (ABP-c)
Interpolated Map Grid (IMG)
Experimental EvaluationConclusion, Ongoing and Future Work
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
10/27
Dept. of Computer Science, Rutgers University
Performance Metrics
Traditional: Distance error between returned and true position
Return avg, 95th percentile, or full CDF
Does not apply to area-based algorithms!
Does not show accuracy-precision tradeoffs!
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
11/27
Dept. of Computer Science, Rutgers University
New Metrics: Accuracy Vs. Precision
Tile Accuracy % true tile is returned
Distance Accuracy distance betweentrue tile and returned tiles (sort anduse percentiles to capture distribution)
Precision size of returned area (e.g.,sq.ft.) or % floor size
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
12/27
Dept. of Computer Science, Rutgers University
Room-Level Metrics
Applications usually operate at the level of rooms
Mapping: divide floor into rooms and map tiles
(Point Room): easy
(Area Room): tricky
Metrics: accuracy-precision
Room Accuracy % true room is the returned roomTop-n rooms Accuracy % true room is among the returnedrooms
Room Precision avg number of returned rooms
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
13/27
Dept. of Computer Science, Rutgers University
1. Simple Point Matching (SPM)
Build a regular grid of tiles, tile expected fingerprint
Find tiles which fall within a thresholdof RSS for each AP
Eager: start from low threshold (= , 2 , )Threshold is picked based on the standard deviation of thereceived signal
Similar to Maximum Likelihood Estimation
=
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
14/27
Dept. of Computer Science, Rutgers University
2. Area-Based Probability (ABP-c)
Build a regular grid of tiles, tile expected fingerprint
Using Bayes rule compute likelihood of an RSS matching thefingerprint for each tile
p(Ti|RSS) p(RSS|Ti) . p(Ti)
Return top tiles bounded by an overall probability that theobject lies in the area (Confidence: user-defined)
Confidence Area size
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
15/27
Dept. of Computer Science, Rutgers University 40
45
50
55
60
65
70
75
80
85
90
x in feet
y
infeet
0 210
140
Measurement At Each Tile Is Expensive!
Interpolated Map Grid: (Surface Fitting)
Goal: Extends original training data to cover the entire floor byderiving an expected fingerprint in each tile
Triangle-based linear interpolation using Delaunay Triangulation
Advantages:
Simple, fast, and efficientInsensitive to the tile size
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
16/27
Dept. of Computer Science, Rutgers University
Impact of Training on IMG
Both location and number of training samples impactaccuracy of the map, and localization performance
Number of samples has an impact, but not strong!
Little difference going from 30-115, no difference using> 115 training samples
Different strategies [Fixed spacing vs. Averagespacing]: as long as samples are uniformly distributedbut not necessarily uniformly spaced methodology hasno measurable effect
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
17/27
Dept. of Computer Science, Rutgers University
Experimental Setup
CoRE802.11 data: 286 fingerprints (rooms + hallways)
50 rooms
200x80 feet
4 Access Points
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
18/27
Dept. of Computer Science, Rutgers University
Area-based Approaches: Accuracy-Precision Tradeoffs
Improving Accuracy worsens Precision (tradeoff)
0 50 100 150 200 25060
65
70
75
80
85
90
95
100
Training data size
%accuracy
Average Overall Room Accuracy
SPM
ABP-50
ABP-75
ABP-95
0 50 100 150 200 2500
2
4
6
8
10
Training data size
%floor
Average Overall Precision
SPM
ABP-50
ABP-75
ABP-95
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
19/27
Dept. of Computer Science, Rutgers University
0 20 40 60 80 1000
0.2
0.4
0.6
0.8
1
probability
ABP-75: Percentiles' CDF
Minimum
25 PercentileMedian75 PercentileMaximum
0 20 40 60 80 1000
0.2
0.4
0.6
0.8
1
probability
ABP-50: Percentiles' CDF
Minimum
25 PercentileMedian75 PercentileMaximum
0 20 40 60 80 1000
0.2
0.4
0.6
0.8
1
distance in feet
probability
SPM: Percentiles' CDF
Minimum25 PercentileMedian75 Percentile
Maximum0 20 40 60 80 100
0
0.2
0.4
0.6
0.8
1
distance in feet
probability
ABP-95: Percentiles' CDF
Minimum25 PercentileMedian75 Percentile
Maximum
A Deeper Look Into Accuracy
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
20/27
Dept. of Computer Science, Rutgers University
Sample Outputs
SPM
ABP-75ABP-50
ABP-95Area expands into the true room
Areas illustrate bias across different dimensions (APs location)
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
21/27
Dept. of Computer Science, Rutgers University
Comparison With Point-based localization:Evaluated Algorithms
RADARReturn the closest fingerprint to the RSS in the trainingset using Euclidean Distance in signal space (R1)
Averaged RADAR (R2), Gridded RADAR (GR)
Highest ProbabilitySimilar to ABP: a typical approach that uses Bayes rule
but returns the highest probability single location (P1)Averaged Highest Probability (P2), Gridded Highest Probability(GP)
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
22/27
Dept. of Computer Science, Rutgers University
Comparison With Point-based Localization:
Performance Metrics
Traditional error along with percentiles CDF for
area-based algorithms (min, median, max)
Room-level accuracy
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
23/27
Dept. of Computer Science, Rutgers University
CDFs for point-based algorithms fall in-between the min, maxCDFs for area-based algorithms
Point-based algorithms perform more or less the same, closelymatching the median CDF of area-based algorithms
Min
Max
Median
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
24/27
Dept. of Computer Science, Rutgers University
Similar top-room accuracyArea-based algorithms are superior at returning multiple rooms,
yielding higher overall room accuracy
If the true room is missed in point-based algorithms the userhas no clue!
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
25/27
Dept. of Computer Science, Rutgers University
Conclusion
Area-based algorithms present users a more intuitive way toreason about localization uncertainty
Novel area-based algorithms and performance metrics
Evaluations showed that qualitatively all the algorithms arequite similar in terms of their accuracy
Area-based approaches however direct users in their searchfor the object by returning an ordered set of likely roomsand illustrate confidence
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
26/27
Dept. of Computer Science, Rutgers University
Ongoing and Future Work
Eiman Elnahrawy, Xiaoyan Li, and Richard P. Martin, The Limits ofLocalization Using Signal Strength: A Comparative Study, In IEEESECON 2004
All algorithms behave the same: there is a fundamental uncertainty
due to environmental effects!
Point-based approaches find the peaks in the PDFs
Area-based approaches explore more of the PDF but cannot narrow it(cannot eliminate the uncertainty)
Useful accuracy, different ways to describe it to users
Future work: additional HW or complex models to improve accuracy
-
7/28/2019 El Nah Raw y 04 Are a Based Presentation
27/27
Dept. of Computer Science, Rutgers University
Thank You