1. The Pothole Patrol: Using a Mobile Sensor Network for Road Surface Monitoring Jakob Eriksson, Lewis Girod, Bret Hull, Ryan Newton, Samuel Madden, Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory

2. Outline Introduction Architecture Data Acquisition Algorithm Performance Related Work Discussion

3. P 2 : A mobile road surface monitoring system Hazardous to drivers and increasing repair costs due to vehicle damage Determine “which” roads need to be fixed Static sensors will not do well – requires mobility! P 2 is first of its kind Challenge : differentiate potholes from other road anomalies (railroad crossings, expansion joints) Challenge : coping with variations in detecting the same pothole. (speed, sensor orientation) P 2 successfully detects most potholes (>90% accuracy on test data)

4. P 2 Architecture Vehicles have GPS and 3-axis accelerometer Opportunistic WiFi/Cellular connections with dPipe to cope network outages Taxi Testbed  7 Toyota Priuses 1  Soekris 4801 2 Embedded Linux  Wifi Card  Sprint EVDO Rev A 3 Network card  GPS Some numerical facts  9730 total kms  2492 distinct kms  7 cabs  174 km with >10 repeated passes 1.http://www.carbuyersnotebook.com/archives/Toyota_Pruis_2006.jpghttp://www.carbuyersnotebook.com/archives/Toyota_Pruis_2006.jpg 2.http://www.pkgbox.org/Soekris-4801.jpghttp://www.pkgbox.org/Soekris-4801.jpg 3.http://gizmodo.com/gadgets/peripherals/two-new-sprint-evdo-rev-a-cards-pantech-px500-and-sierra-wireless-aircard-595-200423.phphttp://gizmodo.com/gadgets/peripherals/two-new-sprint-evdo-rev-a-cards-pantech-px500-and-sierra-wireless-aircard-595-200423.php 1 2 3

5. P 2 Architecture Pothole Record Clustering Cab 1 GPS 3 Axis Accelero meter Location Interpolator Pothole Detector Cab 2 GPS 3 Axis Accelero meter Location Interpolator Pothole Detector Central Server

6. P 2 Architecture Distance Traveled vs. Total Hours Across All Taxis Lower line represents unique roads Segments of roads that were repeatedly covered 258,021 unique road segments

7. DATA ACQUISITION Accelerometer placement  Dashboard  Windshield  Embedded Computer GPS Accuracy  Standard deviation 3.3m

8. DATA ACQUISITION Hand Labeled Data  Smooth Road  Crosswalks/Expansion Joints  Railroad crossing  Potholes  Manholes  Hard Stop  Turn

9. DATA ACQUISITION Loosely Labeled Training Data  We know only types of anomalies and their rough frequencies  Exact numbers and locations are unknown  Extends available training set

10. ALGORITHM Features of accelerometer data High energy events are potholes?  Not really!  Rail road crossings, expansion joints, door slamming are high energy events Accelerometer data is processed by embedded computer  256-sample windows  Pass through 5 different filters

11. ALGORITHM - Filtering Input  Raw accelerometer data  256-sample windows IN Windows of all event classes SpeedHigh-passz-peak xz-ratio speed vs. z ratio OUT Pothole Detections

12. ALGORITHM - Filtering Speed  Car is not moving or moving slowly  Rejects door slam and curb ramp events IN Windows of all event classes SpeedHigh-passz-peak xz-ratio speed vs. z ratio OUT Pothole Detections

13. ALGORITHM - Filtering High-Pass  Removes low-freq components in x and z axes  Filters out events like turning, veering, braking. IN Windows of all event classes SpeedHigh-passz-peak xz-ratio speed vs. z ratio OUT Pothole Detections

14. ALGORITHM - Filtering z-peak  Prime characteristic for significant anomalies  Rejects all windows with absolute z-acceleration < t z IN Windows of all event classes SpeedHigh-passz-peak xz-ratio speed vs. z ratio OUT Pothole Detections

15. ALGORITHM - Filtering xz- ratio  Assumes potholes impact only side of the vehicle  Identifies anomalies that span width of the road (rail crossings, speed bumps)  Rejects all windows with x peak within Δw (=32) samples from z peak < t x X z peak Or, ( X peak / z peak )< t x IN Windows of all event classes SpeedHigh-passz-peak xz-ratio speed vs. z ratio OUT Pothole Detections

16. ALGORITHM - Filtering speed vs. z ratio  At high speeds, small anomalies cause high peak accelerations  Rejects windows where Z peak < t s X speed or, (Z peak /speed ) < t s IN Windows of all event classes SpeedHigh-passz-peak xz-ratio speed vs. z ratio OUT Pothole Detections

17. ALGORITHM – Sample Traces

18. ALGORITHM - Training Tuning parameters t={t z,t x,t s } are computed using exhaustive search over a set of values For each set t, we compute detector score s(t) = corr – incorr 2 Corr is no. of pothole detections when sample was labeled as “pothole” Maximize s(t) Include loosely labeled data s(t) = corr – incorr 2 labeled – max(0,incorr loose – count r )

19. ALGORITHM - Clustering Improve accuracy Cluster of at least k events must happen in the same location with small margin of error(Δd) Clustering algorithm  Place each detection in Δd X Δd grid.  Compute pairwise distances in same or neighboring grid cells  Iteratively merge pairs of distances in order of distance  Max intra cluster distance < Δt  Reported location is the centroid of the locations within it

20. ALGORITHM – Blacklisting & False Negatives Well-known anomalies like bridges, railroad crossings, speed bumps etc can be located from GIS sources and blacklisted GPS errors Pothole avoidance Biased detection will focus on critical anomalies

21. PERFORMANCE EVALUATION Goals  Minimize false negative rate for smooth roads Never a flag a smooth road as anomaly  Missing a few potholes is acceptable Evaluation 1. Classification accuracy on hand-labeled data 2. Performance improvement using loosely labeled data 3. Performance on loosely labeled roads 4. Spot-checks

22. Performance on Labeled Data  Randomly divided into training set and test set  False positive rate is 7.6%  Not accurate PERFORMANCE EVALUATION ClassHand Labeledw/ Loosely Labeled Pothole88.9%92.4% Manhole0.3%0.0% Expansion joints2.7%0.3% Railroad Crossing8.1%7.3%

23. PERFORMANCE EVALUATION Estimating the false-positive rate  Ran the detector on loosely labeled roads  Helps set upper bound on false positive rate (at most 0.15%) on good roads. Road# potholes# windows# detectionsrate Storrow Dr.few186530.16% Memorial Dr.few178120.12% Hwy I-93few287750.17% Binney St.some6887250.63% Beacham St.many164323114%

24. PERFORMANCE EVALUATION Impact of features and thresholds 1. Only Z peak 2. w. xz-ratio filter3. w. speed vs. z ratio t x =1.5t x =2.5 t s =5

25. PERFORMANCE EVALUATION Performance under uncontrolled conditions  Slamming doors  Fiddling with the sensor equipment  Driving behaviors  Deliberately avoiding potholes Use clustering  k=4

26. PERFORMANCE EVALUATION Spot Checks Typical pothole Manhole Expansion joint

27. RELATED WORK  Surveys  Falling weight deflectometer  Machine vision – cameras, robots  Accelerometer  Microsoft Trafficsense – smartphones

28. DISCUSSION This is what I think  Innovative  Ground truth establishment is tedious, expensive in dense road networks  Will it work in hilly areas,slopes?  Future work? Driver feedback – Interactive embedded computers Smartphones – Cheaper solution, greater coverage Comments/Questions ???

29. REFERENCES The Pothole Patrol: Using a Mobile Sensor Network for Road Surface Monitoring Jakob Eriksson, Lewis Girod, Bret Hull, Ryan Newton, Samuel Madden, Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory U. Lee, E. Magistretti, B. Zhou, M. Gerla, P. Bellavista, and A. Corradi. MobEyes: Smart Mobs for Urban Monitoring with a Vehicular Sensor Network. IEEE Wireless Communications, 2006. TrafficSense: Rich Monitoring of Road and Traffic Conditions using Mobile Smartphones Prashanth Mohan, Venkata N. Padmanabhan, and Ramachandran Ramjee {prmohan,padmanab,ramjee}@microsoft.com Microsoft Research India, Bangalore http://research.microsoft.com/apps/pubs/default.aspx?id=70573