Fast spatio-temporal stereo matching for advanced driver assistance systems

This paper presents a new fast spatio-temporal stereo matching approach devoted to Advanced Driver Assistance Systems (ADAS). The proposed method uses dynamic programming (DP) to match edge points of stereo images. The main idea of the paper consists in involving the matching results obtained at a frame in the computation of the disparity map at its following one. At current frame, we propose to compute local and global disparity ranges based on the disparity map of its preceding one. A global disparity range is computed for each image line based on the v-disparity of the preceding frame. If an edge point in the left image at the current frame can be associated with another one in the left image at the preceding frame, a local disparity range is searched for this edge point. Using these disparity ranges in the DP algorithm allows the reduction of the search space as well as the mismatches and speeds up the matching process. The proposed approach has been tested on both real and virtual sequences and the results are satisfactory.