Research Of Moving Target Trajectory Prediction And Obstacle Avoidance In High-speed Scenes Of Intelligent Vehicles

With the increase of car ownership,road traffic accidents are also increasing,and straight roads and general curves of urban roads are frequent scenes of traffic accidents.As one of the solutions to this problem,the research on intelligent vehicle technology is significant and very important.When avoiding other dynamic vehicles,the intelligent vehicle system should first predict the future path of obstacles,and then plan a smooth and traceable collision avoidance trajectory.However,the current kinematic model trajectory prediction method based on physics has a large error and can not meet the accuracy requirements of vehicle collision avoidance.In local path planning,most scholars focus on global path planning or based on known static scenes,while there are few unknown and dynamic researches on actual scenes.In view of the current problems,this thesis focuses on the research of vehicle trajectory prediction and intelligent vehicle path planning and tracking technology under dynamic environment,and designs an active collision avoidance system.The main contents are as follows:(1)The different coordinate systems used by the avoidance system are introduced,and the transformation relationship between the coordinate systems is derived,and the complex problems on different linear roads are converted to the natural coordinate system for simplification;In addition,the detailed characteristics of the dangerous scenarios and distracted driving behaviors studied in this thesis are also introduced.(2)For the trajectory prediction of dynamic vehicles,a state vector is constructed to describe the relationship between vehicles and roads,and three driving intention models including left lane change,right lane change and lane keeping are established by using Hidden Markov Model(HMM)to identify the driving intention of dynamic vehicles.In addition,Long Short Term Memory(LSTM)network is used to predict the vehicle trajectory in the future period of time under different intentions.In the experiment,the influence of different input step sizes of different LSTM models on the prediction accuracy is analyzed,and finally the appropriate input step size of each model is determined.The root mean square error of prediction is verified.(3)A local path planning method is proposed to decouple the vehicle transverse and longitudinal.The lateral displacement and longitudinal velocity are optimized by quadratic programming algorithm in the corresponding convex space.Through comparative experiment,the influence of each weight value of quadratic programming on the planning results is analyzed.With the progress of the experiment,each weight value is gradually determined,and finally a smooth and traceable collision avoidance path and relatively stable planning speed are obtained.The tracking error model is established according to the two-degree-of-freedom dynamic model of the vehicle.The horizontal LQR controller and the longitudinal double PID controller with feedforward control are designed.(4)The obstacle avoidance process under several typical working conditions was verified in the Carsim/Simulink/Prescan simulation environment,and the trajectory planning and speed planning results of self-vehicle under different curvatures were analyzed.