Early Obstacle Detection Using Region Segmentation and Model Based Edge Grouping

| This paper describes a system able to detect obstacles as soon as they appear on the horizon by the analysis of a sequence of images recorded by a forward looking TV camera mounted on a vehicle. It exploits several suggestions coming from the a priori knowledge of the scene where the vehicle is moving. The evaluation of the luminance differences between obstacle and road surface and the correlation of motion in succeeding images are the main tools used to formulate obstacle hypotheses. A dynamic model based matching algorithm performs the recognition and the validation of the obstacle.