Identification of pedestrian motion feature in mixed traffic conditions and anti-collision algorithm of autonomous vehicle

Abstract

To improve the safety of pedestrians crossing the road in the mixed traffic conditions, this study proposed an active collision avoidance method based on the prediction of pedestrian trajectory. A convolutional neural network is applied to identify the motion feature of pedestrians crossing the road in the image of an automated driving environment with vehicular sensors. Then combined with the pedestrian motion parameters, a new Kalman filtering algorithm is proposed to analyze the change of pedestrian motion feature and predict the trajectory of the pedestrian. Furthermore, a PDS (Pedestrian, Distance, and Speed) estimated braking distance model based on pedestrian characteristics, the distance between pedestrian and vehicle, and the speed of the vehicle is established in this study for the autonomous vehicle controlling speed in advance to avoid risks. It improves both the crossing road pedestrian safety and efficiency of traffic. Eventually, simulations based on CarSim/Simulink are designed to verify the validity of the method. Results show that the method proposed can effectively predict pedestrian trajectories and realize active collision avoidance under the time delay of the detection link.