Abnormal Behavior Recognition Of Expressway Vehicles Based On Vision Detection And Localization

Nowadays,the traffic monitoring coverage of urban expressway is getting higher and higher,and more and more monitoring data streams are introduced into the central control room at the same time.How to quickly and accurately locate the vehicles with abnormal behaviors in the numerous monitoring videos poses a lot of challenges to the backstage personnel on duty.With the help of computer vision,the main research content of this paper is to detect the vehicles in the monitoring data stream,identify the abnormal behavior,and assist the backstage personnel to judge.Combined with the actual needs of urban expressway application scenarios,this paper designs a vehicle orientation point detection algorithm,which is used to solve the problem that vehicle trajectory deviates from the real lane of the vehicle;at the same time,this paper proposes a vehicle abnormal behavior recognition algorithm based on trajectory information,which is used to judge whether the vehicle has abnormal parking,illegal U-turn,illegal Lane occupation and speeding.The specific research results are as follows:(1)An algorithm for detecting the orientation point of vehicle is proposed.When the vehicle is located at the edge of the monitoring field and deviates from the camera’s main optical axis,the trajectory drawn on the road plane by using the vehicle geometric center point will deviate from the lane where the vehicle is located.In view of this problem,this paper creatively proposes the vehicle orientation point,the orientation point annotation tool is developed for the construction of dataset.In the test dataset,the average absolute error of the end coefficient fitting is 0.1289,and the average deviation degree is smaller.Therefore,the algorithm can accurately fit the position of the vehicle orientation point.(2)A vehicle abnormal behavior recognition algorithm based on trajectory information is designed.In this paper,the physical information of vehicle trajectory is restored by distortion correction,and the finite state automata with hidden state is designed to judge the abnormal behavior of vehicle logically.The test results show that the time segment detection accuracy of the algorithm is 86.12%in the experiment of abnormal parking.While in the experiment of illegal lane occupation,the accuracy of lane information recognition is 95.52%.And in the experiment of illegal U-turn,the recall rate of vehicle driving direction detection is 93.16%.Finally,In the experiments related to overspeed detection,the average absolute error of speed measurement is 1.4544m/s,which proves that the delay trigger mechanism can promote the improvement of the detection accuracy of the system.(3)This paper develops a vehicle abnormal behavior recognition system under the scene of expressway.The system can not only display the input video stream in real time on the monitoring display windows,but also visualize the vehicle trajectory extraction module and abnormal behavior recognition module.At the same time,the system has the function of abnormal vehicle license retention,which directly stores the vehicle snapshot of abnormal vehicle.Based on the actual needs of urban expressway application scenarios,this paper designs trajectory extraction algorithm based on orientation point detection and abnormal behavior recognition algorithm based on vehicle trajectory,and a system platform is built to integrate the algorithm model.Through the practice in Jingtong expressway project,it is found that the work of this paper meets the predetermined expectations and has academic and application value.