1. Nericell: Rich Road and Traffic Monitoring using Mobile Smartphones
Prashanth Mohan
Venkat Padmanabhan
Ram Ramjee
Microsoft Research India, Bangalore
Presented by
Ali Khodaei

2. Traffic Monitoring Developed Countries Developing Countries
GPS based tracking is adequate
Infrastructure support exists
Developing Countries
Bangalore ≠ <your favorite US city>

3. What’s Different? Potholes Road bumps Varied vehicle types
Liberal honking
Chaotic intersections …

4. Why Rich Monitoring?
Find least stressful route

5. Widespread distribution of mobile phones
Road and Traffic Monitoring
Without deployed infrastructure
Using existing mass of mobile phones

6. Mobile Phones ~3 billion phones worldwide ~300 million phones in India
115 million of 1 billion phones sold worldwide in 2007 were smartphones
7 million added in India each month

7. Mobile Phones Mobility Computing + communication + sensing
Far more capable & ubiquitous than special purpose sensors

8. Nericell Overview Using sensors individually
Accelerometer
drive quality
Microphone
honking
GSM radio and GPS
localization
Using sensors in combination
Virtual reorientation of accelerometer
Distinguishing pedestrians from stop-and-go traffic
Triggered sensing

9. Energy Challenge
Measurements made on HP iPAQ hw6965 9

10. Accelerometer-based Sensing
Advantage: low energy cost
Challenge: “disorientation”
Analyses:
braking detection
bump/pothole detection
pedestrian versus stop-and-go traffic

11. Braking Detection Braking impacts drive quality Two approaches:
GPS: high energy cost (600 mW on iPAQ hw6965)
Accelerometer: much cheaper (2 mW + 30mW)
Accelerometer-based braking detection:

12. Braking Detection
braking would cause a surge in aX because the accelerometer would experience a force pushing it to the front.
To detect the incidence of braking
compute the mean of aX over a sliding window N seconds wide.
If the mean exceeds a threshold T, that event is declared as a braking event

13. Virtual Reorientation

14. Virtual Reorientation
Euler Angles:
Any orientation of the accelerometer can be represented by Z-Y-Z (and other equivalent) rotations
Three Unknowns (angles):
pre-rotation (φpre ),
pre tilt (θtilt),
post-rotation (ψpost)
Knowns:
Gravity along Z
Braking along X

15. Euler Angles

16. Virtual ReorientationUsing Gravity

17. Using Braking

18. Automatic Virtual Reorientation

19. Results: Virtual Reorientation

20. Braking detection with Virtual Reorientation

21. Differentiating pedestrians from stop-and-go traffic

22. Road bump detection

23. Locating a Road Bump Accelerometer is cheap, so keep on continuously
When bump is detected, use GSM tower matching to estimate location
median error: 130 m, 90th percentile: 610 m
Send bump report to server
If several reports in same vicinity, server triggers GPS on other phones for location fix
Sample result: GPS turned on only 3.2% of the time on a 20 km drive with one point of interest

24. Pothole Detection

25. Pothole Detection High speed (≥ 25 kmph)
z-peak: look for significant spike

26. Pothole Detection

27. Pothole Detection Low speed (< 25 kmph)
z-sus: look for sustained dip

28. Results: Pothole Detection
Training data: 5km long drive with 44 bumps
Test data: 35km long drive with 101 bumps

29. Microphone-based Sensing
Advantage: ubiquity
Challenge: energy, privacy
Analyses:
honk detection: triggered when accelerometer indicates a lot of braking
vehicle type: exposed versus enclosed vehicle

30. Honk Detection Efficient detector suitable for mobiles
discrete Fourier transform on 100 ms of audio
look for spikes in the kHz range
spikes at least T times the mean, where T ranges between 5 and 10
Performance: 5.8% of CPU on the HP iPAQ

31. Honk Detection

32. Triggered Sensing
Use cheap sensors to trigger the activation of expensive sensors when needed
Examples:
Virtual Reorientation: accelerometer -> GPS
Localization: GSM -> GPS
Honk detection: accelerometer -> audio

33. Resources
Trafficsense: Rich monitoring of road and traffic conditions using mobile smartphones
P Mohan, VN Padmanabhan, R Ramjee - SenSys'08

34. Questions?!