Vehicle Trajectory Prediction Based On Driving Intention Recognition

Since Google announced the start of the deployment of autonomous driving technology in 2009,autonomous driving technology has been highly valued by governments and enterprises in various countries.Since 2019,my country has proactively laid out its autonomous driving development strategy and issued a series of relevant policy documents to vigorously drive the development of autonomous driving technology.Different from the development route of bicycle intelligence in European and American countries,my country combines the powerful advantages of new infrastructure and uses the coordinated development of vehicles and roads as the basis to realize the implementation of autonomous driving technology.However,the realization of the popularization of autonomous driving is still very far away,and human-machine hybrid driving will exist for a long time.Starting from the development trend of autonomous driving,this paper analyzes the main safety issues faced by autonomous driving technology,combined with the characteristics of automatic driving data acquisition,and proposes the importance of predicting the trajectory of the vehicle in front of the autonomous driving car to improve safety.Establish a prediction model for the trajectory of the preceding vehicle based on the recognition of driving behavior intentions,use the self-driving car to obtain information and powerful processing capabilities beyond the visual range,use historical trajectory data as the main feature information,predict the future trajectory of the preceding vehicle,and perform the prediction results Evaluation,which provides a meaningful reference for the planning and decision-making research of autonomous vehicles.The research content and results of this article mainly include the following aspects:(1)Analyzed the difference between automatic driving and manual driving from the perspective of perception and decision-making,and pointed out that in the context of human-machine hybrid driving,irregular manual driving behavior is a key problem in predicting the trajectory of the vehicle in front,mainly involving lane changing and turning scenarios.A more comprehensive survey of the existing autonomous driving data sets was carried out,and the Waymo and NGSIM data sets were analyzed.It was concluded that most of the autonomous driving data sets were oriented to visual recognition and multi-source sensor fusion technology,and lacked sufficient traffic flow information.Considering the applicability of research traffic scenarios,the NGSIM data set is selected and preprocessed.(2)This article combines the working mechanism of autonomous vehicles and the characteristics of autonomous vehicle information acquisition beyond visual range under vehicle-road collaboration,and builds a forward vehicle trajectory prediction model based on long-and short-term neural network LSTM,and defines multiple types according to the difference in feature information The study found that in addition to the basic position information,adding the feature information speed difference and headway in the historical trajectory can significantly increase the actual effect of the vehicle trajectory prediction model;adding interactive feature information can make the model prediction effect more Best fit the actual road conditions.The results show that the TC-LSTM model proposed in this paper performs better in the evaluation index.(3)Aiming at the lack of interaction between vehicles and the lack of consideration of driving behavior in the TC-LSTM model,a vehicle trajectory prediction DNN-LSTM model based on driving intention recognition is proposed.The DNN network is used to identify the driving behavior intention,and the driving intention is used as a feature input,combined with the historical trajectory information of the preceding vehicle,to predict the trajectory of the preceding vehicle.The research obtains the relationship between the number of LSTM neurons in the DNN-LSTM model and the model effect,and analyzes the prediction effect of the model under different prediction time domains.The results show that the DNN-LSTM prediction model proposed in this paper can predict the future trajectory of the preceding vehicle within 4s within a small error range.