Research On Active Collision Avoidance Pedestrian System Based On Visual Perception

As an significant part of vehicle active safety technology,vehicle active collision avoidance pedestrian system has been a hot point of advanced assisted driving system.As pedestrians are important urban traffic participants,how to make vehicles avoid pedestrians autonomously and safely is a serious problem of vehicle active collision avoidance pedestrian system.Based on the vehicle camera sensing pedestrians,this dissertation proposes an active collision avoidance pedestrian system,which integrating longitudinal and lateral active collision avoidance control.The following problems are analyzed:(i)pedestrian collision risk assessment and executing collision avoidance mode,(ii)horizontal active collision avoidance path planning,(iii)longitudinal and horizontal active collision avoidance vehicle control,(iv)simulation working condition.(1)The pedestrian information captured by the vehicle camera is continuously detected by the YOLOv4 target detection network from the actual road condition ahead sensed by the vehicle camera in real time.And then according to the pedestrian detection results,the pedestrian distance is estimated based on the geometric characteristics.(2)In order to assess the collision risk of human-vehicle more accurately and reduce the impact of pedestrian distance estimation errors,a Kalman filter algorithm is designed to estimate the state of pedestrian motion.Based on the TTC model,a safety vehicle distance model is introduced as the collision risk assessment model in this dissertation.Finally,the longitudinal and lateral active collision avoidance decision mechanisms are designed.(3)Longitudinal and lateral active collision avoidance controllers are designed.First,the vehicle dynamics model is built as the control object of the collision avoidance controller.A two layer longitudinal active collision avoidance controller is designed,the upper layer controller adopts fuzzy control theory to output the desired deceleration of the vehicle according to the relative distance and relative speed between the vehicle and the target pedestrian;the lower layer controller adopts PID algorithm to adjust the actual deceleration of the vehicle according to the difference between the desired deceleration and the actual deceleration of the vehicle.So that it can control the vehicle to accurately track the desired deceleration.The lateral active collision avoidance controller is also designed as upper and lower layers.Under the premise of collision avoidance path planning principle,the upper layer controller adopts the quintic polynomial algorithm to design a reasonable lateral collision avoidance path;the lower layer controller adopts the model prediction control algorithm.Due to the high real-time requirement of controller,the linear vehicle prediction model is derived.And the vehicle lateral collision avoidance path tracking is also analyzed under multiple constraints.With reference to the C-NCAP test standard for pedestrian collision avoidance,the longitudinal and lateral active collision avoidance system test conditions are designed respectively.And virtual traffic scenarios are built in the Prescan software.The effectiveness of the proposed pedestrian active collision avoidance system is verified by joint simulation.