Far Away Obstacle Detection by Perceptual Integration in Traffic Scenes

Abstract

The detection of obstacles is a critical task of computer vision systems devoted to driving assistance of vehicles moving on structured roads. This paper proposes an algorithm for the early detection and tracking of vehicles based on perception of the main structures localized in correspondence of the road in the image sequence. Perceptual persistence of some structures is used to start up object hypotheses, then a model based grouping process integrates these initial hypotheses along the sequence until a real obstacle is identified by the description of its main structures.