Car-following Behavior Analysis Of Drivers Based On Natural Driving And Dangerous State Recognition

The rapid increase in the number of motor vehicles and the number of drivers has had a great impact on people’s production and life,and problems such as traffic congestion,traffic accidents and environmental pollution have also become more and more serious.Car-following behavior of drivers is an important part of driver’s behavior.Research on driver’s car-following behavior can improve the safety of drivers’ driving behavior,minimize the occurrence of rear-end collisions,and reduce the total number of traffic accidents.The exit section of the expressway is a place where accidents frequently occur.In order to improve the safety of the traffic flow at the exit of the expressway,this article collects data related to the driver’s car-following behavior at the exit of the expressway to establish a safety database and a risk database for the driver’s car-following behavior.The forest model recognizes the driver’s dangerous car-following state.First,use DJI drones to shoot high-definition traffic flow videos,collect natural driving data,and extract vehicle motion trajectories through Tracker software,including: the relative distance,relative speed,relative acceleration,and headway of each vehicle to the preceding vehicle.,Establish the driver’s car-following behavior feature matrix,and extract the driver’s car-following fragment data set according to the driver’s car-following fragment extraction criteria,extracting a total of 89 car-following fragments in 1 lane,extracting a total of 423car-following fragments were extracted.The car-following behavior characteristic parameters of the driver at the vehicle layer and the traffic layer are selected,and a hierarchical linear model is established.The analysis finds that the traffic flow information has a nested and hierarchical data structure.The factors that affect the driver’s car-following behavior come from the vehicle layer and the car-following behavior.There are two levels of traffic flow,which is different from the traditional theory of driver’s understanding of carfollowing behavior.Then,based on the traffic safety evaluation index TTC and considering the driver’s situational awareness characteristics,the criteria for determining the driver’s dangerous car-following state are established,and the driver’s car-following behavior database is divided into a driver’s car-following state safety library and a dangerous library.The analysis found that the data of the driver’s car-following state hazard database only accounts for 1.2% of the total driver’s car-following data,and there is an imbalance problem in the data set.This paper combines K-means clustering and SMOTE algorithm to solve the imbalance problem of the driver’s car-following state data set.Use the K-means algorithm to cluster the samples of the driver’s dangerous car-following state,record the cluster center position of the clustering result,and apply the SMOTE algorithm to the data set to increase the sample operation at the cluster center position until the target imbalance rate is reached.Through the independent sample T test,it is found that there is no significant difference between the data set after the KM-SMOTE algorithm and the original data set.Through Gaussian distribution fitting,it is found that the distribution of the original data set and the optimized data set are the same,indicating that the structure and distribution of the dangerous sample data set after processing by the KM-SMOTE algorithm have not changed.Finally,a car-following state recognition model is established based on the random forest,the importance of the driver’s car-following behavior characteristics is evaluated,and the grid search method is adopted to determine the optimal parameter combination of the random forest model.And compare and analyze the classification results of the random forest model of the original data set of driver-following behavior,the classification result of the random forest model of the data set processed by the SMOTE algorithm,and the classification result of the random forest model of the data set processed by the KM-SMOTE algorithm.The comparison results show that the classification results of the random forest model of the data set processed by the KM-SMOTE algorithm are all the best.The KMSMOTE algorithm can improve the classification performance of the random forest model and can effectively carry out the dangerous state of the driver.The judgment.