Vehicles Detection in Stereo Vision Based on Disparity Map Segmentation and Objects Classification

This paper presents a coarse to fine approach of on-road vehicles detection and distance estimation based on the disparity map segmentation supervised by stereo vision. Scene segmentation is first performed relying on the robustness of the UV-disparity maps to generate free space and obstacles space. This last is investigated for on-road vehicles detection. The detection process starts with off-road objects substraction based on the connected component labeling algorithm which is also used for on-road segments extraction instead of the traditional hough transform for more robust, precise and fast detection. Objects classification is then applied to the on-road segments by using some cues describing the geometry of vehicles like width and height. However, these latter have been measured not in meter but rather in pixels in function of the disparity. The whole approach is presented and the experimental results of evaluation are shown.