Vehicle Behavior Recognition and Prediction Method for Intelligent Driving in Highway Scene

Vehicle behavior in high-speed dynamic driving scenarios is highly interactive, but much information in the surrounding driving environment cannot be directly observed. Intelligent driving vehicles need to have the ability to recognize and predict the behavior of surrounding vehicles in order to make reasonable decision planning. First, a vehicle behavior recognition method based on GMM-HMM is established in this paper. The probability of its possible behavior is estimated through continuous vehicle state observations. Then, this paper constructs a Bayesian network vehicle motion prediction model, considering the interactive information of the target vehicle and the environmental vehicles, and estimate the probability distribution of vehicle trajectories based on different vehicle behaviors. Furthermore, use the NGSIM data set based on real highway scene information to train, test and verify the model. Finally, research results show that the vehicle behavior recognition model has high sensitivity and accuracy, and the accuracy of lane change recognition before 1 second can reach more than 95%. The prediction model can predict unknown samples more stably without overfitting problems. The difference between the overall prediction accuracy of the training set and the test set is controlled within 2.7%, and the input of interactive information has a significant impact on the prediction of vehicle behavior.