Stereo vision based pedestrian detection using B-spline modeling

This paper presents a new approach for pedestrian detection in the vehicle environment using stereo vision. The segmented pedestrians, represented as a set of equal disparity attributed segments, are refined using B-spline arithmetic. These segments are obtained from a non-dense disparity map. The disparity map is estimated using a new developed algorithm relying on the contour-following algorithm. The approximated segments are then matched with preexisting pedestrian model. The pedestrian model is generated by utilizing B-spline functions, which can refine the human body contour extracted in a digitized image. Some attributes of the approximated segments like concavity, convexity, and local curvature are considered to perform the matching process.