Research On Vehicle Collision Avoidance Strategy Based On Pedestrian Crossing Intention And Trajectory Prediction

At present,with the increase of car ownership and the rapid development of autonomous driving technology,vehicle active collision avoidance system have developed rapidly with the promotion of automotive "new four modernizations".As a vulnerable group among traffic participants,pedestrians are more vulnerable to harm in traffic scenarios.Pedestrian safety issues have become one of the severe challenges facing the world.Related research is also committed to improving the safety of pedestrians in traffic scenarios.At present,although there are many studies of pedestrians’ collision control,there are few research on the subjective street intention of pedestrians and the trajectory of the street forecasting,and the considering collision control control scenarios are relatively single,and there are certain limitations.To this end,this article studies a pedestrian-based street intention prediction algorithm based on characteristic fusion,and proposes a trajectory prediction algorithm that integrates pedestrian cross-street intentions.Based on this,a pedestrian active collision control algorithm is designed to improve road pedestrians to participate in traffic.Safety.The main research work of the paper is as follows:(1)In order to obtain high-accuracy and rich number of pedestrian cross-street data sets,pedestrian cross-street data collection experiments have been conducted.The advantages of millimeter wave radar and single camera have been designed to design data fusion algorithms based on BP neural networks.Through the spatial and temporal fusion of the two sensor data,the fused data increased by 2.77% and 6.62% respectively in the longitudinal and transverse distance compared with the measured data of a single sensor,Improved pedestrian detection accuracy.(2)In order to make up for the deficiency of the existing pedestrian intention prediction model that only considers the single skeleton characteristics of pedestrians,and accurately judge the pedestrian intention to cross the street,first of all,analyze the four common types of pedestrian crossing movement feature changes,and use Tiny-YOLOv4 and Deep Sort algorithms to achieve pedestrian detection and tracking;Then,based on the attention model,the pedestrian head orientation feature is integrated into the pedestrian skeleton feature to improve the expression of pedestrian movement feature,and then establish a pedestrian crossing intention prediction model based on feature fusion;Finally,the model performance is verified in the actual traffic scene,and compared with other intention prediction models.The results showed that the accuracy and F1 scores of pedestrian intention prediction model based on feature fusion increased by 5.66% and5.9% respectively compared with the intention prediction model only considering single skeleton features.(3)In order to eliminate the impact of pedestrian space-time interaction and accumulated errors in trajectory prediction,and improve the accuracy of trajectory prediction,a trajectory prediction model integrating pedestrian crossing intention and historical trajectory is established based on pedestrian kinematics characteristics and LSTM trajectory prediction model,introducing speed prediction module,attention mechanism and environmental feature extraction module.The pedestrian space-time interaction network based on adaptive graph convolutional network and a pedestrian feature fusion algorithm based on attention model are created;Finally,the pedestrian trajectory prediction experiments under four kinds of crossing intentions are carried out and compared with the existing trajectory prediction network.The results show that compared with STGCNN network,the trajectory prediction model with crossing intentions proposed in this paper has lower ADE and FDE in KITTI data set and self-made data set.(4)In order to verify the active algorithm performance of vehicles based on pedestrian prediction trajectories,first divide the vehicle’s vertical and horizontal collision area based on the TTC risk assessment model;It has formulated a variety of collision scenes.3 types of collision avoidance strategies and designed the corresponding collision control algorithm.Based on Carsim,Prescan,and Simulink,set up a joint simulation platform,and made 3 active collision algorithms to simulation and verification.Finally,conducted real vehicle experiments in different collision avoidance scenarios.Simulation validation and real vehicle test results show that the vehicle collision avoidance algorithm based on pedestrian crossing intention and trajectory prediction can achieve vehicle prediction of the intention and trajectory of pedestrians on the road in multiple scenarios,effectively avoiding pedestrians,and the collision avoidance effect meets the design requirements.The research has important research value for improving the safety of pedestrians participating and improving the active collision avoidance performance of vehicles.