Research On Behavior Recognition And Trajectory Prediction For Lane Changing Based On Internet Of Vehicles

China’s car ownership has increased year by year,making the traffic environment increasingly complex,caused by the driver’s lane change operation,traffic safety issues have aroused widespread concern.With the development of intelligent transportation system,the emergence of car network technology can be real-time transmission and feedback of data generated during vehicle driving,in which the car-to-vehicle communication technology can be transmitted between vehicles.The development of 5G communication technology in China has provided technical support for the vehicle network and also provided a new perspective for the study of vehicle rerouting.In this paper,we will identify the changing behavior of a vehicle and predict its changing trajectory based on real data generated during the vehicle’s travel.The findings of this paper will contribute to the safety of vehicle lane changing in a connected vehicle environment and reduce traffic conflicts caused by lane changing.Traditional vehicle lane changing models are mostly based on mathematical as well as physical models that do not adequately take into account the uncertainty and randomness of driver lane changing behavior in increasingly complex road environments,and the data for training models are mostly based on data generated by driving simulators rather than by vehicles traveling under real road conditions.To address these problems,this paper proposes a neural network lane changing model based on the real data of the driver’s driving,which can identify the lane changing behavior and predict the lane changing trajectory through an in-depth analysis of the vehicle lane changing data.The main research elements of the article are as follows.(1)Staging and analysis of the vehicle’s lane changing process identifies the factors that characterize lane changing behavior.Characteristic data is extracted from data from the SPMD(Safety Pilot Model Development)project to characterize driver lane changing behavior.(2)A model for the recognition of commutation behavior is constructed based on the good nonlinear fitting ability of the Feedforward Neural Network(FNN),and the model is trained and tuned by SPMD data to determine the optimal network structure of FNN.The trained FNN model was subjected to lane switching behavior recognition and the corresponding recognition rate under the optimal time window reached 90.7%.The ROC curve checksum model shows that the FNN model satisfies the requirements for the identification of changing lane vehicles in a networked vehicle environment.(3)Long Short-Term Memory(LSTM)neural network is built to predict the vehicle’s changeover trajectory,and the model is trained and tested by 300 sets of changeover trajectory samples selected from SPMD data,and the mean absolute error and root mean square error are used as reference indicators for model tuning.The LSTM model has a small mean error for the lateral trajectory of the changing vehicle and for the longitudinal position,which is more accurate and less error than the dynamic neural network prediction.