Development of Pedestrian Collision Avoidance System Based on Crosswalk Recognition by an On-Board Camera

Abstract

This paper proposes a collision warning system based on crosswalk detection and pedestrian detection by using a monocular camera to prevent crashes with pedestrians. First, the paper describes a crosswalk detection algorithm which includes the edge extraction process and the cross ratio feature calculation. Next, based on the position of the detected crosswalk, the paper proposes two pedestrian detection algorithms, one is based on the object motion and the other one based on the crosswalk brightness pattern. The performance of each algorithm is evaluated from the viewpoint of computation time and recognition rate. Finally, A visual and auditory warning system based on detection result is presented and its feasibility is examined by test drives by the experimental vehicle.