Pedestrian-vehicular collision avoidance based on vision system

Vehicle-pedestrian collision avoidance is an important issue on intelligent vehicles research field. Although pedestrian detection and collision avoidance were concentrated by previously many studies, there are many problems about the collision prediction and warning. Based on pedestrian detection, the trajectory prediction and time to collision (TTC) still needs to be further studied to prevent vehicle-pedestrian collision. In this study, a monocular camera with low cost was utilized. The well performance of pedestrian detection was shown. Based on this detection system, a model of pedestrian trajectory prediction was proposed. Then, the extended Kalman filter was used to predict motion of pedestrian. Based on trajectory prediction, time to collision range (TTCR) was proposed identify three levels risks, including safe, potential collision and collision. In this study, the on-road experiments were conducted to validate the proposed method. The experimental result shows that the TTCR is more reasonable and safe to apply in vehicle-pedestrian collision avoidance.