Vehicle Trajectory States Identification On The Freeway Entrances And Exits

Freeway entrances and exits are frequent accident areas in expressway sections and their safety affects the capacity and service level of expressways.Studying the vehicle trajectories in the entrance and exit area can be helpful to understand the vehicle trajectory behaviors at the entrance and exit,realize the vehicle risk warning under machine vision and improve the entrance and exit traffic safety.This paper focuses on the state identification of vehicle trajectories at the entrance and exit of the expressway.The specific research work is as follows:(1)Vehicle identification and tracking.In traditional vehicle recognition methods During identification state,in order to solve the problem of vehicle recognition difficulty caused by incomplete background modeling,a three-frame difference and improved Gaussian mixture model method with Canny edge detector was used to separate the vehicle from the background.During the tracking state,Kalman filters and an improved iterative Hungarian algorithm were used to predict and match vehicle positions in the video.The results show that compared with other background modeling methods,the method in this paper has better recognition ability for vehicles in the video,and the improved tracking algorithm has stronger tracking ability and fewer iterations.The selected video segment is used to detect the vehicle tracking effect after modeling,and the average accuracy rate is 93.56%.(2)Identified vehicle trajectories cluster using hierarchical spectral clustering.Firstly,the trajectories were fitted by the least square method and divided into straight line clusters and curve clusters according to the curvature of the trajectories.Second,trajectory angle similarity was used as a parameter to cluster the vehicle trajectories using spectral clustering.Then the velocity and acceleration speed parameters were used to perform secondary spectral clustering on the trajectories.The results show that the method mentioned above can distinguish the lane keeping trajectories from lane change trajectories,and the method also has a certain degree of discrimination for the second lane change trajectories.The trajectories can be further distinguished by clustering the velocity and acceleration speed parameters.(3)Establish vehicle trajectory states model based on CHMM+RF theory.the CHMM+RF model was trained with the training set and tested with the test set.The model was cross-verified with the CHMM model and CHMM+SVM model.The results show that the CHMM+RF model has a higher accuracy than other models.In this paper,the trajectory video of vehicles on freeway entrance and exit is used.The trajectories data were extracted by identifying and tracking the vehicles in the video.the CHMM+RF model was used to identify the trajectory states.The identification method used in this paper has higher accuracy,which can be used as a reference for vehicle state identification of freeway entrances and exits and other road sections with machine vision.