Lane-Changing Decision Intention Understanding Model Of Surrounding Drivers For Intelligent Driving

With the development and gradual application of intelligent driving technology,in the future,the Automatic Vehicles(AVs)and artificial vehicles will drive together.Due to the AVs could not understand the driving intention of surrounding artificial vehicles accurately,so that the AVs may make irrational behavior decisions.This may produce negative impact on the traffic operation safety.In order to solve this problem,it is necessary to predict the driving intention of the surrounding vehicles effectively,especially the Lane-Changing Decision(LCD)intention,so as to optimize the control strategy of AVs.The formation mechanism of driver’s LCD intention is complex and influenced by many factors,such as,driver’s psychology and driving style.For lacking the consideration of driver’s psychology and driving style,the existing studies may not be able to predict the LCD intention accurately varying with the drivers and traffic conditions.Aiming at the above problems,this paper proposes a LCD intention understanding model of surrounding drivers oriented to intelligent driving,which integrates the driver’s psychology and driving style in the modeling process.In the construction process of the model,firstly,starting from the driver’s own perception characteristics,the psychological stimulation of the surrounding traffic environment on the driver is abstracted as the “field” model.Then combining the driver’s visual attention range with visual attention distribution in different driving situations,the driver’s focus area could be defined.And the quantitative method of driver’s psychological stress under the influence of dynamic traffic environment could be constructed,which could provide technical support for the construction of LCD model.Secondly,a hierarchical driving style recognition framework is proposed,which considers the influence of driver’s state on driving style,including driving style offline training module and online inference module.The offline training module utilizes the historical vehicle trajectory data and Gaussian Mixed Model(GMM)clustering to classify the drivers,and constructs the layer-by-layer mapping relationship among driving behavior data,driving state,and driving style.The online inference module uses the real-time vehicle trajectory data and Support Vector Machine(SVM)algorithm to judge the driver’s state and style online.Finally,coupling driver’s psychological state and driving style,a LCD intention understanding model is constructed.And the Light Gradient Boosting Machine(Light GBM)algorithm from decision tree classifier is used to learn and predict the LCD intention.In the validation of the model,the I-80 highway vehicle trajectory data from Next Generation Simulation(NGSIM)database are utilized to verify the effectiveness of driving style hierarchical identification model and the LCD intention model proposed in this paper.The experimental results showed that,after considering the influence of driving state on driving style,the average maximum identification accuracy of driving style could reach95.92%.Therefore,the hierarchical driving style identification model proposed in this paper could identify driver’s driving style effectively.For the LCD model,after adding driver’s psychology and driving style,the average prediction accuracy of LCD model was 97.21%.Compared with other common LCD models,the accuracy of LCD prediction has been improved effectively.In the intelligent driving environment,the model could predict the LCD intention of surrounding vehicles,and then provide support for the decision-making of AVs.