The present paper proposes a computer vision algorithm which detects obstacles (namely vehicles) moving ahead of a vehicle, equipped with a forward looking TV camera, while it runs on highways. The algorithm is based on the evaluation of the edge segments, which are detected within a restricted area of interest located on the lane where the monitored vehicle moves. Geometrical and perceptual grouping of horizontal and vertical segments is compared with a simple model of vehicles and, exploiting time integration, reaches successful recognition in few images even if the obstacles appear far away near the horizon line and are merged with a cluttered background. Some results are presented using sequences of images recorded on highways near Torino. For the covering abstract see ITRD E114174.
Samenvatting