Billboard sweep stereo for obstacle detection in road scenes

An algorithm for stereo-based obstacle detection in a general road scene is proposed. In the proposed algorithm, the vertical obstacles on a horizontal ground surface are defined as billboard planes, and the distant environment is modelled using a background wall. During stereo matching, each image pixel has a disparity value which is the distance to the billboard or the slope of the ground surface. The algorithm finally obtains a set of the billboard planes and efficiently manages them using frame coherence. Experimental results show that the use of billboard planes can effectively reduce the search range of disparities. Also, the proposed method produces reliable range data in accordance the KITTI benchmark.