Driving Style Recognition Method Based On Traffic Scenes

With the development of autonomous driving technology,which promotes the steady progress of intelligent and networked vehicles,people have higher requirements for autonomous driving.In autonomous driving,the synergy between the advanced driver assistance system and the driver is an important part.The personalized difference of drivers’ tendentious behavior during driving is defined as driving style.Accurate identification of driving style is of great significance for designing ADAS more in line with drivers’ expectations and improving road traffic safety.In the existing researches on driving style,most of the data sources are collected by driving simulation system,which is different from the real driving data.Moreover,most studies only focus on a single working condition or a specific scene,and the scenarios considered are not comprehensive enough.Considering the above problems,the real road test data set is selected to analyze the characteristics of different driving styles in each typical traffic scenario,and the differences between scenarios are compared to model the driving styles in multiple scenarios.The main research contents are as follows:(1)Considering the influence of traffic scenes on driving behavior,the distribution characteristics of driving behavior parameters of vehicles in various scenes are studied and the comprehensive characteristic parameter sequence that can characterize driving style is obtained.Firstly,three typical traffic scenarios were divided based on NGSIM data set: ramp entry scenario,lane change scenario and car following scenario.Constraints such as vehicle location and speed were used to screen vehicle track data samples in each scenario.Secondly,the distribution characteristics of driving behavior parameters such as vehicle speed and time headway in each typical traffic scenario are analyzed,which is used as the basis for the selection of driving style characteristic parameters.Finally,factor analysis method is used to reduce the dimension of the feature parameters to obtain the comprehensive feature parameters that can characterize the driving style.(2)Research on classification and recognition of driving styles based on typical traffic scenarios.The K-means++ algorithm was used to cluster the comprehensive feature parameters in ramp entry,lane change and car following scenarios to obtain driving style categories.By analyzing the characteristic parameter distribution of each driving style,it is found that there are obvious differences among different driving styles,such as speed,lateral speed,time headway,etc.At the same time,by comparing the distribution of characteristic parameters of vehicles in different traffic scenes,it is found that the effective characteristic parameters in scene A cannot represent the driving style in scene B.(3)Machine learning algorithm is used to model driving style.By combining the traditional Tr-training with cross entropy,the difference between base classifiers is expanded to reduce the noise samples generated in the training process.The driving style recognition model was constructed based on different traffic scenarios,and the experimental results showed that the accuracy of the constructed recognition model reached more than 93% in the three scenarios.(4)Modeling the driving style with the driver model.Considering the comprehensive influence of transverse and longitudinal dynamics,a driver model of transverse and longitudinal comprehensive control is established.MPC is used to control the yaw motion characteristics of the vehicle,PID is used to follow the longitudinal speed of the vehicle,and the input of the simulated vehicle is constrained by the driving style characteristic parameters obtained above,so as to realize the simulation of different driving styles under different scenarios.