Research On Trajectory Prediction And Collision Warning Strategy For Right-turning Vehicles And Pedestrians At Intersections

As one of the key active safety technologies for vehicles,the forward collision warning system is facing increasingly stringent performance requirements with the continuous development of autonomous driving technology.Pedestrians,as an important group of traffic participants,are one of the important protection targets in the field of active vehicle safety.To address the problem that existing active collision warning systems do not fully consider the motion states of vehicles and pedestrians,a graded collision warning strategy based on vehicle-pedestrian trajectory prediction is designed to provide a safer collision warning strategy for autonomous vehicles.The right-turn scenario at intersections with frequent conflicts between pedestrians and vehicles is selected for studying the collision warning method between right-turn vehicles and pedestrians ahead.The main research contents include the following aspects:(1)Collection of trajectory data for right-turn vehicles and pedestrians at intersections.Typical urban road intersections are selected to collect trajectory data for right-turn vehicles and pedestrians crossing the street.The raw data is preprocessed,and the behavioral characteristics and variation patterns of right-turn vehicles and pedestrians are analyzed and summarized.The areas where collisions with pedestrians are likely to occur are analyzed based on the variation patterns of right-turn vehicle speed,providing a basis for building trajectory prediction models and setting simulation parameters in the following sections.(2)Trajectory prediction for right-turn vehicles and pedestrians.Correlation analysis is performed on the collected research data to provide a basis for selecting feature parameters for trajectory prediction.Attention-LSTM neural network model is employed to predict the trajectories of right-turn vehicles entering and leaving the intersection,considering the time series characteristics of the trajectory.RMSE and ADE are used to evaluate the accuracy of trajectory prediction.Experimental results show that the model has high prediction accuracy within 4 seconds of prediction time.Based on the motion law of pedestrians,a pedestrian kinematic model is constructed,and a Kalman filter with adaptive noise update is used for real-time trajectory prediction of pedestrians.The experimental results show that the predicted trajectory matches the actual trajectory well,with an average displacement error of 0.169 m and high accuracy of prediction results.(3)Construction of a graded collision warning model for pedestrians in front of rightturning vehicles.First,a vehicle lateral risk area model was constructed,and based on the analysis of vehicle and pedestrian trajectories,the future lateral position of pedestrians and the longitudinal position of vehicles were estimated.A lateral and longitudinal collision risk judgment model based on vehicle and pedestrian trajectory prediction was designed.Second,a collision time-to-contact(TTC)and braking distance model was established for pedestrians with collision risk targets.By integrating collision time-to-contact,escape time-to-arrival,braking distance,and relative distance,a graded warning and braking strategy was designed.Finally,based on actual research data,a right-turn vehicle and pedestrian collision simulation scenario was constructed in the Prescan/Simulink joint simulation environment to verify the effectiveness of the graded collision warning strategy under different driving conditions.Simulation results show that under different driving conditions,the strategy can provide early warning and apply comfortable deceleration during the braking process to ensure pedestrian safety while considering comfort.