Research On Vehicle Trajectory Prediction And Lane Change Intention Recognition Method Based On Transformer

With the development of autonomous driving technology,vehicle trajectory prediction and lane-change intention recognition have become important issues in autonomous driving systems.These tasks require modeling and prediction of vehicle trajectories,as well as analysis and judgment of vehicle behavior and intentions.Lane changes may increase the risk of traffic accidents,so autonomous driving technology needs to identify vehicle lane change intentions and predict future trajectories,and calculate the safe distance of surrounding vehicles to reduce risks and reduce lane changes.Therefore,developing accurate trajectory prediction models is crucial to improving lanechange safety.Currently,deep learning technology has become the main method for processing vehicle trajectories and lane-change intentions,and many deep learning methods based on the NGSIM and Argoverse vehicle trajectory datasets have been developed for trajectory prediction and lane change intention recognition.This paper improves the Transformer model and designs new model structures for predicting vehicle driving trajectories and recognizing vehicle lane change intentions on the Argoverse and NGSIM datasets,respectively.In the task of predicting vehicle driving trajectories,this paper compares and evaluates the model with various trajectory prediction methods using metrics such as minimum average displacement error(minADE),minimum final displacement error(minFDE),and miss rate(MR)on the Argoverse dataset.The results show that the Transformer-based model improved in this paper has higher accuracy in the trajectory prediction task.In addition,this paper further improves the Transformer model to enable it to perform vehicle lane change intention recognition tasks.The paper also conducts comparative experiments with LSTM networks with different hidden layer numbers,and the results show that the Transformerbased vehicle lane change intention recognition model in this paper is superior in terms of accuracy and recall rate.The experimental results of this paper demonstrate that the improved Transformer model used in this paper can effectively predict vehicle trajectories and recognize lane change intentions,with high accuracy and robustness.Compared with the LSTM model,it has the advantages of faster speed and higher accuracy,and can provide strong support for the implementation of future autonomous driving systems.