Development of Pedestrian Collision Warning System Based on Vehicle Dynamics Estimation by Optical Flow

Abstract

This research aims to develop a collision warning system for preventing vehicle-pedestrian accidents by making use of the information of crosswalk position. The system is based on an assumption that all moving objects near a crosswalk are inferred as pedestrians. Crosswalks can be detected by a monocular camera. Moving objects are detected by a monocular camera with an optical flow algorithm. In the proposed pedestrian detection algorithm, precise localization of crosswalk is necessary for searching pedestrians near a crosswalk and warning the driver of the collision risk. Consequently, this paper proposes a vehicle egomotion estimation method by the optical flow and the method of correcting the distance to crosswalk, pedestrian position and velocity during dynamic movement of a car, such as pitching angle of the vehicle. Test drives show the effectiveness of the proposed method on the accuracy of the distance to crosswalk, the pedestrian position and the pedestrian velocity.