The Research On Prediction Of Driver’s Lane Changing Intention Based On LSTM-CNN

Autonomous vehicles are of great significance to reduce traffic congestion and traffic accidents.However,the sudden lane change behavior of vehicles in adjacent lanes poses a serious threat to the driving safety of autonomous vehicles.This requires autonomous vehicles to have the ability to predict the driving intentions of other drivers in advance.Therefore,this thesis proposes a method for predicting driver’s lane changing intention based on Long ShortTerm Memory(LSTM)neural network and Convolutional Neural Network(CNN):First,the reasons for the driver’s lane-changing intention and the lane-changing decisionmaking process are analyzed.Every stage of lane changing process has characteristics that can reflect the driver’s intention of lane changing.The features of lane change intention generation stage and lane change preparation stage can reflect the driver’s lane change intention earlier,but these features are often not reliable enough.The features of lane change execution stage have high reliability,but the prediction range of lane change intention using these features is very little.In this thesis,a mixed structure neural network based on lstm-cnn is proposed to predict the driver’s lane changing intention.Secondly,the data source of this thesis is the NGSIM natural vehicle trajectory data set,but the original data in the NGSIM data set contains wrong data and needs to be processed.This thesis uses a symmetric exponential moving average filter(sEMA)to smooth the original data.Then the lane changing trajectory data of vehicles is extracted,and the trajectory data of surrounding vehicles at each time of lane changing vehicles is extracted.The speed difference and longitudinal distance between lane changing vehicles and their surrounding vehicles are calculated.These data reflect the traffic environment around lane changing vehicles.Finally,the neural network structure of the driver’s lane change intention prediction model is designed.The prediction model consists of two parts,LSTM part and one-dimensional CNN part.After training,the prediction accuracy of the prediction model at 3 s,2s,1s,and 0s before changing lanes reach 76.77%,85.35%,93.95%,and 94.44%,respectively.Compared with SVM,Decision tree,LSTM and CNN alone,the results show that the prediction accuracy of the driver’s lane change intention prediction method proposed in this thesis is significantly better than other methods involved in the comparison.The thesis uses the NGSIM data set and deep learning algorithms to deeply study the driver’s lane-changing behavior and the driver’s decision-making process in the lane-changing behavior.Through the analysis of lane-changing behavior and decision-making process,the characteristics that can be used to predict the driver’s lane-changing intention at each stage of lane-changing are selected,and the driver’s lane-changing intention prediction model is established.Related researches in the thesis provide a theoretical basis for improving the driving safety and ride comfort of autonomous vehicles.