Title: The Pothole Patrol: Using a Mobile Sensor Network for Road Surface Monitoring
1The Pothole Patrol Using a Mobile Sensor Network
forRoad Surface Monitoring
- Jakob Eriksson, Lewis Girod, Bret Hull,
- Ryan Newton, Samuel Madden, Hari Balakrishnan
- MIT Computer Science and Artificial Intelligence
Laboratory - Mobisys 2008
- Speaker Lawrence
2Outline
- Introduction
- P2 Architecture
- Data Acquisition
- Algorithm
- Evaluation
- Conclusion
- QA
3Outline
- Introduction
- P2 Architecture
- Data Acquisition
- Algorithm
- Evaluation
- Conclusion
- QA
4Introduction
- Motivation
- Maintain roadways spend millions of dollars, but
few feel comfortable. - There are cause of expensive lawsuits and claim.
5Introduction (cont.)
- Goal
- Detecting and reporting the surface condition of
roads.
6Introduction (cont.)
- Pothole Patrol (P2 )
- Deploy P2 on 7 taxis running in the Boston.
- P2 uses the inherent mobility of the
participating vehicles. - Opportunistically gather data from accelerometer
and GPS sensors. - Process the data by using machine learning to
assess road surface condition.
7Introduction (cont.)
Accelerometer GPS
TAXI
7 cabs are able to cover 2492 distinct kilometers
during their normal driving in 10 days
8Outline
- Introduction
- P2 Architecture
- Data Acquisition
- Algorithm
- Evaluation
- Conclusion
- QA
9P2 Architecture (cont.)
Testbed Soekris 4801 embedded computer(Linux)
WiFi card Network card GPS 3-axis
accelerometer
Soekris 4801
Placement (Consider by accelerometer ) Firmly
attached to dashboard inside cars glove box
10P2 Architecture
Client (Cars)
Server
11P2 Architecture (cont.)
Combine Filter
Sensing
121E31 25N02
121E31 25N02
Upload Store
Final Report
Clustering
12P2 Architecture (cont.)
13Outline
- Introduction
- P2 Architecture
- Data Acquisition
- Algorithm
- Evaluation
- Conclusion
- QA
14Data Acquisition
- The distribution of coverage density across all
the lengths of road encountered.
15Data Acquisition (cont.)
Good
16Data Acquisition (cont.)
- GPS Accuracy
- GPS accuracy in our deployment is important.
- Measure accuracy
- Placed a thick metal bar across a road, and
repeatedly drove over it. - standard deviation of the positions reported for
the bar to be 3.3 meters.
17Data Acquisition (cont.)
- Hand-labeled Training Data
- Collect the data by repeatedly driving down
several known stretches of road in the Boston.
Event Class
18Data Acquisition (cont.)
- Loosely Labeled Training Data
- Hand-labeled less coverage non-pothole type
- With types and rough frequency of anomalies
- Without exact number and location
Example
19Outline
- Introduction
- P2 Architecture
- Data Acquisition
- Algorithm
- Evaluation
- Conclusion
- QA
20Algorithm
- Main Concept Various road conditions introduce
high z-axis acceleration. - It is less sufficient to only use accelerometer
to identify the real pothole! - Use filter to reject non-pothole even type.
21Filtering stage
22 Filter
Smaller highway anomalies
Xlttx Z
Speed vs Z ratio
OUT pothole detection
ts
Expansion joint ,rail crossings
ZlttS V
XZ-ratio
IN windows of all event type
Z-peak
tz
Minor anomalies
Zlttz
23Tuning the parameter
- Goal minimizing false positive rate
- Corr is the number of correct detections of
pothole - Incorr is the number of incorrect detection of
pothole (false positive)
t tz ,tx ,ts
s(t)corr-incorr2
Arg max s(t) t
minimizing false positive rate
24Outline
- Introduction
- P2 Architecture
- Data Acquisition
- Algorithm
- Evaluation
- Conclusion
- QA
25Performance of labeled data
Class before after
Pothole 88.9 92.4
Manhole 0.3 0.3
Exp. Joint 2.7 0.0
Railroad Crossing 8.1 7.3
Test data of listed class that was reported as
potholes by our algorithm, before and after
training on additional loosely labeled data.
False positive rate 7.6
26Impact of features and thresholds
Z-peark filter only
Z-peak, xz ratio
Z-peak, xz ratio, speed vs z ratio
27Performance on uncontrolled cab data
- Using 7 taxis running through the Boston
downtown to collect data - Clustering the detections that requiring 4 trace
before reporting them as a detection
2848 detections with cluster size 4
29Outline
- Introduction
- P2 Architecture
- Data Acquisition
- Algorithm
- Evaluation
- Conclusion
- QA
30Conclusion
- Studied an application of mobile sensing
detecting and reporting the surface conditions of
roads. - By using GPS and accelerometer, the P2 system is
able to detect adverse road conditions. - The false positive rate is very low in
uncontrolled taxis experiment.
31QA