Road surface distress detection in disparity space

Automatic identification of potholes on roads is needed to avoid traffic accidents or to ensure better driving comfort. This paper presents a strategy which focuses on the detection of major distress present on a road surface using stereo vision. The proposed strategy performs road-plane modelling directly in the image-disparity space, without back-projecting a disparity image into 3D space. The modelling is based on a RANSAC process that finds the dominating plane for locating potholes being below surface level. We also formulate an alternative implementation using the v-disparity technique. Both implementations are evaluated on an urban dataset which indicates a very high accuracy of our dominating-plane method in terms of pothole detection.