Sensys 2009 Speaker:Lawrence. Introduction Overview & Challenges Algorithm Travel Time Estimation Evaluation Conclusion.

VTrack: Accurate, Energy-Aware Road Traffic Delay Estimation Using Mobile Phones

Sensys 2009

Speaker:Lawrence

Outline

Introduction Overview & Challenges Algorithm Travel Time Estimation Evaluation Conclusion

Outline

Introduction Overview & Challenges Algorithm Travel Time Estimation Evaluation Conclusion

Introdution

Motivation Traffic delays and congestions Real time traffic information

Challenges Energy consumption Inaccurate position samples

VTrack Vehicles as probes  A real time traffic monitoring system

Motivating Problem How the quality of VTrack’s travel time estimates on the sensor

being sampled and the sampling frequency.

Introdution

Key finding HHM-based map matching is robust to noise

Travel times estimated from WiFi localization alone are accurate enough for route planning

Travel times estimated from WiFi localization alone cannot detect hotspots accurately

Sampling GPS periodically to save power

Introdution

Contribution Quantitative evaluation of the end to

end quality of time estimates from noisy and sparsely sampled locations.

Outline

Introduction Overview & Challenges Algorithm Travel Time Estimation Evaluation Conclusion

System Overview

Key Application Detecting and visualizing hotspots Real time route planning

iPhone web page

Requirements

Accuracy For route planning , errors in the 10%~15%

range.

Efficient enough to run in real time Some existing map-matching algorithm run A*

style shortest path algorithm

Energy efficient GPS excessively drains the battery

Challenges

Map matching with outages and errors.

Time estimation - even with accurate trajectories is difficult

Localization accuracy is at odd with energy consumption

Outline

Introduction Overview & Challenges Algorithm Travel Time Estimation Evaluation Conclusion

Algorithm

HMM A Markov process with a set of hidden

states and observables.

Viterbi Decoding Dynamic programming tech Find the maximum likelihood sequence

of hidden states given a set of observables and emission probability and transition probability.

HMM

Hidden state: road segments Observables: position samples Transition probability: from one road to next Emission probability: conditional probability of

<segment, position>

Match mapping process

1 2 3 4

Outline

Introduction Overview & Challenges Algorithm Travel Time Estimation Evaluation Conclusion

Travel Time Estimation

The traversal time T(s) for any segment S:

Estimation Errors Outages during transition times.▪ Intersection delay

Noisy position samples▪ Noisy sensor

Outline

Introduction Overview & Challenges Algorithm Travel Time Estimation Evaluation Conclusion

Data collection

Raw data 800 hours 25 cars

Evaluation of Route Planning

WiFi good enough

Evaluation of Hotspot Detection

Detect 80%~90% of hotspots. Not too aggressive.

Evaluation of Energy Accuracy

Estimating WiFi Cost The cost per sample of GPS is 24.9X the cost per sample

of WiFi. 8% of total power consumption

Offline Energy Optimization (Assuming the WiFi cost is 1 unit)

Impact of Noise

Outline

Introduction Overview & Challenges Algorithm Travel Time Estimation Evaluation Conclusion

Conclusion

Using mobile phones to accuracy estimate travel times using inaccurate samples.

Address key challenge 1. reducing energy consumption 2. accurate travel time from inaccurate rate

positions

VTrack uses an HMM-based map matching scheme.

Successfully identify highly delayed segments and accuracy route planning with noisy.