A study on pedestrian detection models based on real accident data from IVAC database in Changsha of China

This study aims to evaluate the probabilities of pedestrian detection within the response time of vehicle collision avoidance system of a car. For this purpose, the carpedestrian accident scenarios were analyzed using selected data from the IVAC accident database, in which the cases were collected from in depth investigations of the accidents in Changsha of China. The selection criteria were: (1) the accident occurred between 2001 and 2008; (2) the accident involved a passenger car, SUV, MPV or pick-up truck; (3) the pedestrian was not standing still before impact. Based on these criteria, 389 car-pedestrian cases were selected. The two most common scenarios (F1 and F2) were identified as the pedestrian crossing a straight road from the left (F1) or the right (F2) of the drivers. A mathematical model was developed with the frontal impact cases of F1 or F2 scenario. The following four parameters describing the configuration before the accident were studied: the trajectory and speed for both the car and the pedestrian. Considering the different half detective angles of the sensor system (15 degree, 30 degree, 45 degree), the probabilities of pedestrian detection were calculated. It was found that when the half detective angle was equal or larger than 30 degrees the sensor system could detect more than 94% of the pedestrians in both evaluated scenarios