Stereo vision enabling fast estimation of free space on traffic roads for autonomous navigation

A novel algorithm capable of estimating free space for vehicle navigation is presented. When a disparity map — dense or sparse — from stereo matching and a longitudinal profile of the road surface on the disparity domain are provided, the free space is estimated precisely. According to the longitudinal profile of the road surface, the disparity map is classified into an obstacle disparity map and a road surface disparity map. After combining these two disparity maps through a score map, a border line separating the road surface and the non-road surface is estimated using dynamic programming on a udisparity representation. The main contribution of the proposed approach is the robust detection of the free space and the distance between stereo cameras and obstacles, whereby the detection is sufficiently rapid for vehicle navigation. The validity of the proposed algorithm is demonstrated by experiments through many outdoor road images from various traffic scenarios.