Road Pothole Detection System Based on Stereo Vision

In this paper, we propose a stereo vision system which detects potholes during driving. The objective is to benefit drivers to react to potholes in advance. This system contains two USB cameras taking photo simultaneously. We use parameters obtained from camera calibration with checkerboard to calculate the disparity map. 2-dimensional image points can be projected to 3-dimensional world points using the disparity map. With all the 3-dimensional points, we use the bi-square weighted robust least-squares approximation for road surface fitting. All points below the road surface model can be detected as pothole region. The size and depth of each pothole can be obtained as well. The experiments we conducted show robust detection of potholes in different road and light conditions.