Vehicle Detection And Tracking Based On Virtual Traffic Data Set And Its Application

Abundant data resources in traffic scenarios provide a rich data foundation for vehicle detection,tracking and application research.Multi-target detection and tracking in related technologies mainly implements vehicle tracking,positioning and classification functions,so as to accurately obtain road traffic operation conditions,and make a significant contribution to the handling of traffic emergencies and road traffic construction planning.According to the current development of the virtual engine technology,in view of the limitations of the open multi-target tracking data set in real traffic scenes,this paper analyzes and constructs a virtual traffic data set.At the same time,researches on the acquisition of multi-target tracking and smooth trajectory,and further analyzes and realizes the vehicle parameter extraction and event detection.The main research contents are as follows:1.In view of the shortcomings of public multi-target tracking data sets in real traffic scenarios,a multi-target tracking data set was constructed based on virtual traffic scenarios.The data set was driven by the Unreal Engine 4 virtual engine as the bottom layer,and the operation of road traffic vehicles was truly restored on the Carla simulation platform.The virtual traffic data set had the advantages of huge data volume,complex scenes,and detailed data file information.Finally,according to the virtual traffic data set,it analyzed its applications such as simulating abnormal vehicle events,three-dimensional detection,and speed estimation.2.Research on multi-target tracking technology in traffic scenes.After obtaining the accurate detection results of the YOLOv4 target detection algorithm,combining the target detection results and the scale adaptive filter trained by the target itself,and then according to the variance of the target response value when the occlusion occurs and the Kalman linear uniform velocity model,realize an improved single target tracking algorithm based on KCF.Single target tracking algorithm,which could obtain a more fitting target bounding box and dealt with missed detection and occlusion.In order to improve the matching accuracy of the multi-target detection frame and the existing trajectory,combining multiple feature measurement methods,a multi-feature fusion data correlation multi-target tracking algorithm was proposed to achieve continuous tracking and trajectory acquisition of multiple targets in the scene.Finally,aiming at the problem of trajectory jitter,the Savitzky-Golay trajectory smoothing algorithm was proposed to obtain a smooth trajectory of the target.3.Vehicle parameter extraction and event detection in traffic scenarios.Firstly,according to the smooth trajectory of the vehicle obtained by the multi-target tracking algorithm,the vehicle’s travel direction and the up and down detection lines were automatically set to accurately obtain multiple road traffic parameters.Secondly,by fusing scattered trajectory information in the scene,a method for constructing,maintaining and updating road traffic patterns based on voting mechanism was proposed to enhance the initiative of the algorithm,and thus it realized the detection of individual abnormal events of vehicles.Finally,the algorithm proposed in this paper was tested and analyzed in real road traffic scenes and virtual road traffic scenes,and the feasibility and accuracy of the algorithm were verified,indicating that the method used in this paper can meet the needs of practical applications,and promotes the development of intelligent transportation.