Research On Intention And Trajectory Prediction Method Of Intelligent Vehicle Under Lane Changing Scene

At present,intelligent vehicles have become the focus of academia and industry.As the key technology of intelligent vehicles,intelligent vehicles need to accurately predict the behavior and movement of the surrounding vehicles and other obstacles in the traffic scene.The decision-making system makes the best decision based on motion prediction information,so as to avoid dangerous scenes and ensure safe and comfortable ride experience.Intention and trajectory prediction can be collectively referred to as motion prediction and vehicle motion prediction is to predict the future movement of vehicles based on their own historical trajectory information and traffic environment.As the actual driving scene is often complex and changeable,accurate motion prediction needs to consider the interaction between vehicles and model the driving target.In addition,intelligent vehicles need to consider the evolution and development of future traffic,and update the motion prediction results in real time according to the changes of environmental information.The comprehensive consideration of vehicle intention and motion state can significantly improve the prediction reliability.In the open and complex environment,how to make vehicles have the ability of autonomous prediction and reasoning like human drivers has become a research hotspot.The development of probability graph model and other methods provides an effective technical way to solve this problem.The main contents of this paper are as follows:1.The intention prediction based on Dynamic Bayesian network is carried out,and the influencing factors of lane changing behavior are analyzed.The driving behavior related factors are selected as prediction variables,and the variables are discretized.By constructing a three-layer Dynamic Bayesian network,the driving intention reasoning model is established,and the joint distribution of variables is expressed.The perceived physical information is used as evidence for driving intention reasoning,and the uncertainty is characterized by probability.2.The vehicle behavior is represented as hidden variable,and the behavior is identified based on Hidden Markov model and Gaussian Mixture model.Considering the evolution of future traffic situation,this paper calculates the revenue through the front drivable space,comfort and safety,and uses the maximum utility theory to infer the future behavior.The final intention prediction result is obtained by combining the former two.This method can make up for the shortcomings that dynamic Bayesian network can’t be used for continuous variables and has poor interpretability.The NGSIM data set is extracted,and the extracted data is used to learn the model parameters.3.Based on the Frenet framework,the trajectory is generated with the lane change duration as the variable,and the optimal trajectory is selected by the loss function which chooses the maximum lateral acceleration and lane change duration as the indexes.At the same time,based on the current motion parameters of the vehicle,the trajectory is predicted through the vehicle kinematics model.The final trajectory prediction result is obtained by combining the two methods.The prediction results not only have high accuracy in the short term,but also conform to the vehicle movement mode in the long term.