An obstacle detection method based on binocular stereovision

Abstract

As the main tasks of Advance Driver Assistance Systems (ADAS), obstacle detection has attracted extensive attention. Traditional obstacle detection methods on the basis of monocular vision will lose its effect when new obstacles appear or the obstacles have severe occlusion and deformation, so this paper proposes an obstacle detection method based on disparity map, which can detect all obstacles on the road accurately. We first determine the disparity of the road in V disparity map through an approach based on weighted least square method. Then we obtain the disparity map contains only the obstacles on the road, and generate corresponding Real U disparity map by projection. Finally, obstacles are detected in Real U disparity map. Experiments show that the proposed method can not only precisely detect the obstacles at greater distances and the obstacles with a large area of occlusion, but also accurately calculate the distance information according to the disparity of the obstacle.