Research On Lane-changing Behavior Recognition And Trajectory Prediction When Drivers Intervene In Manmachine Cooperation Mode

Man-machine co-driving is the only way for the long-term development of intelligent vehicles.In this mode,intelligent vehicles cooperate with drivers to complete the driving task.In order to effectively reduce the decision-making conflict between the intelligent system and the driver when the driver is involved,it is necessary to accurately identify and predict the driver’s behavior and expected trajectory.This paper identifies the different lane-changing behaviors when the driver is involved and predicts the driver’s expected lane-changing trajectory according to the current driving state and the surrounding environment information,so as to provide support for man-machine cooperation hybrid intelligent decision-making.The main contents of this thesis are as follows:1.The lane-changing scene is divided into lane-changing freely and lane-changing with certain risk.Based on the driving simulator,the behavior data acquisition test when the driver intervenes in the system and the lane-changing trajectory data acquisition test when the driver’s driving are carried out respectively under the man-machine cooperation mode.The purpose of this is to provide data support for the establishment of the driver’s lane-changing behavior recognition model when the driver intervenes in the system and the driver’s expected lanechanging trajectory prediction model under the man-machine co-driving mode.2.Based on the experimental data collected by the driver’s intervention behavior under the man-machine cooperation mode,the characteristics of the driver’s operation data under different lane-changing behaviors are analyzed and the driving behavior recognition model is established.The lane-changing behavior of drivers during intervention is divided into three categories: lane keeping,free lane-changing and lane-changing with certain risk,and the lanechanging behavior recognition models are established based on support vector machine and bidirectional long-short term memory network respectively.Considering the recognition accuracy and recognition efficiency,the optimal time window,network structure and related parameters of the model are determined.The test results show that the average prediction accuracy of the Bi-LSTM-based behavior recognition model is 92.1%,which is 2.89% higher than that of the SVM model,indicating that the deep learning model has more advantages than traditional machine learning methods in the performance of behavior recognition.3.Based on the lane change trajectory acquisition test data collected under the driver’s operation,the differences between different lane changing behavior trajectories are analyzed,and the prediction models of drivers’ expected lane changing trajectories are established based on Bi-LSTM network and Bi-GRU network respectively.The historical trajectory characteristics of free lane change and lane change with certain risks are deeply mined.The vehicle driving state parameters and the surrounding vehicle information are used to predict the position of the vehicle in the future.The root mean square error is selected as the evaluation index of the model.The test results show that the prediction error of bi-directional recurrent neural network is smaller than that of unidirectional recurrent neural network,and the performance of Bi-GRU trajectory prediction model is the best and converges faster than Bi-LSTM.Finally,the Bi-LSTM behavior recognition model and the Bi-GRU lane change expected trajectory prediction model are connected through UDP communication,and the test set data are used to verify the real-time performance and accuracy of the algorithm.