The Pothole Patrol (P2) uses the opportunistic mobility of sensor-equipped vehicles to detect and report the surface conditions of roads. Each car in the system carries a CarTel node with 3-axis acceleration and GPS sensors, gathering location-tagged vibration data at a frequency of 400 Hz (we would've liked to use a higher sample rate, but the sensors we use max out at 400 samples per second).
P2 uses CarTel's opportunistic wireless protocols to deliver the data over whatever wireless network is available to a back-end server. The server processes this vibration data using signal processing and data correlation techniques to assess surface conditions.
We have deployed P2 on 10 taxis (out of the 27 in our fleet testbed) running in the Boston area. Our analysis algorithms, which we calibrated against a human's perception of how bad a given pothole or surface is, are able to detect 75% of surface conditions reported as bad by a human observer, with a false positive rate of less than 5%. Overall, in a week's worth of driving, our system found about 4,800 bad surface locations.
We went out and took pictures of some of the highest scoring detected problems. You can see why these things make you grimace when you drive over them. A schematic of the P2 data processing pipeline is shown in the picture on the right, below.
Below is a map overlay showing some of the automatically discovered “potholes”. Several of these are real potholes, but some of them aren't (though they cause jolts or unseemly vibrations). These “false positives” are generally caused by manholes, drains, tiled roads, and such like.
Click on the red icons below to see more information about each detection.
Jakob Eriksson, Lewis Girod, Bret Hull, Hari Balakrishnan, and Sam Madden.