A novel, signal model based approach to lane detection for use in intersection assistance

Detection of marked intersections poses challenging demands on the quality of a lane marking detection. Therefore, a novel approach to single lane marking segment detection is proposed that enforces the detected objects to be of rectangular shape. Two algorithms are given to incorporate this knowledge into the detection process. The first algorithm minimizes the squared error between the grey value image and a two dimensional grey value model of the lane marking segment. The second algorithm adaptively binarizes image portions and then fits a total least squares line to the pixel positions in the world. Test results on real imagery are presented. It is found that the integration of this knowledge considerably improves detection results, when compared with a generic edge detector