V Track: Energy-Aware Traffic Delay Estimation Using Mobile Phones PowerPoint Presentation, PPT - DocSlides

VTrack: Energy-Aware Traffic Delay Estimation Using Mobile Phones

Lenin Ravindranath, Arvind Thiagarajan, Katrina LaCurts, Sivan Toledo, Jacob Eriksson, Sam Madden, Hari Balakrishnan

Massachusetts Institute of Technology

Motivation

Traffic applicationsReal time traffic congestion informationRoute planning - traffic aware routingTraffic delay prediction

Traffic delays

and congestion

Wasted fuel

Commuter frustration

4.2 billion hours in 2007 spent struck in traffic

Estimate current delay on each road segment

Vtrack Goal

Route planning

Hot spot detection

Road segment delay estimates

Approaches

Flow monitoring sensorsHigh deployment costGPS equipped probe vehiclesCover large areasDeployment costEnd user smart phonesLarge penetration and massive amount of data

Sensors: GPS, Wi-Fi, GSMOn roads and time useful for other users

Challenges

Inaccuracy

of position samplesEnergy

consumption

GSM

GPS

Wi-Fi50m

200m

5m

VTrack

Wi-Fi

Infrequent GPS samples

Wi-Fi localization

War driving: Access point - GPS mappingAP observations -> Centroid location

Noise

Outliers

Outages

Delay estimation

Map matching

- Sequence of segments

Find delay on road segments

Map matching

Hidden Markov Model

S1

S2

S3

p1

p2

p3

p4

S1

S2

S3

1/3

1/3

1/3

S1

S2

S3

S1

S2

S3

S1

S2

S3

S1

S2

S3

p1

p2

p3

p4

Viterbi

Noise

- Gaussian

Outliers

- Speed constraint

Outages

- Interpolation

Dealing with outages

Delay on segments

S1

S2

S3

p1

p2

p3

p4

p1

p2

p3

p4

S1

S1

S3

S3

T (S1) = t(p2) – t(p1)

+ ½ (t(p3) – t (p2))

T (S3) = t(p4) – t(p3)

+ ½ (t(p3) – t (p2))

VTrack

ApplicationsRoute PlanningShortest time path between a source and a destinationHotspot detectionFinding road segments that are highly congested

Evaluation

Analyzed

over 800 hours of drive data

25 cars with both GPS and Wi-Fi

Key Results

HMM based map matching is robust to noiseTrajectories with median error less than 10%Delay estimates from Wi-Fi are accurate enough for route planningThough individual segment delay estimates have 25% median errorOver 90% of shortest paths have travel times within 15% of true shortest path

Accurately detect over 80% hotspots with less than 5% false positives

Further workSampling GPS infrequently

Improves the accuracy of Wi-Fi based estimatesAnalyzed energy consumptionAdaptive samplingDynamically selects best sensorBased on road networks, accuracy, energySegment delay prediction