Research On Dynamic Vehicle Trajectory Prediction Method In Lanes

In order to avoid the collision between the intelligent driving vehicle and other manual driving vehicles during road driving,it is necessary to predict the driving trajectory of the manual driving vehicle in the future for a certain period of time,so that the intelligent vehicle can make reasonable decision-making planning and improve driving safety and riding comfort.However,most of the current methods for predicting vehicle trajectories remain at the theoretical research stage based on models and data drives.They have shortcomings such as short prediction time,single processing scene,poor real-time performance and stability,and cannot be put into practical use.Therefore,this paper designs a lane sequence prediction algorithm based on long short-term memory(LSTM)and develops a real-time dynamic vehicle future trajectory prediction system.The system consists of four modules: container,analyzer,evaluator and predictor.The container is responsible for data storage and preprocessing.Firstly,it uses the Kalman filter algorithm to process the sensed surrounding vehicle information to improve the credibility of the upstream sensed information;and then constructs the lane sequence of each obstacle vehicle in the future through positioning information and high definition map information to prepare for the lane sequence evaluation;finally extracts planning information to provide data support for scene analysis.The analyzer is responsible for handling complex conditions during the prediction process.It uses the scene-dividing method to deal with the complex problems of the road scene;it adopts the method of dividing the right of way to solve the game problem of vehicles at the junction;at the same time,it prioritizes the obstacle vehicles,directly ignore the vehicles that do not affect the driving of intelligent vehicle,and improve the real-time performance of the system.The evaluator is responsible for evaluating the probability of each lane sequence.For different scenarios,the cruise LSTM evaluation model and the junction LSTM evaluation model are proposed based on the timing characteristics of the long short-term memory network.Through experiments,the accuracy rate of the two sequence evaluation models for lane sequence probabilities reached more than 90%.The predictor is responsible for generating the future trajectory of the vehicle.Including lane sequence predictor,movement sequence predictor and junction predictor for different scenarios.The actual vehicle experiments in highway and urban road environments showed that the three predictors could predict the vehicle’s driving trajectory in 8s at a frequency of 10 Hz per second and the average error could be controlled at about 5m.In summary,the research technical route of this thesis is correct,the experimental data is true and reliable,the trajectory prediction accuracy,real-time performance and stability have been significantly improved,the prediction effect is satisfactory,and it has high practical value and theoretical reference significance.