Multi Target Recognition And Tracking For Vehicle Abnormal Behavior

With the rapid development of China’s science and technology and economy,China’s car ownership is still increasing significantly.However,a series of road safety problems such as traffic congestion and traffic accidents are also on the rise.Among them,most of the tragedies caused by abnormal vehicle behaviors are caused by drunk driving or fatigue driving,which are fatal injuries to themselves or others,Therefore,vehicle abnormal behavior recognition,as an important key technology in the field of intelligent traffic safety management and intelligent city traffic management,has become one of the important contents of today’s artificial intelligence research and application.In recent years,traffic supervision technology mainly focuses on the detection of vehicle violations,but there are still deficiencies in collision accidents caused by abnormal vehicle behavior and early warning.The traditional means of traffic supervision is based on the naked eye.It not only costs a lot of manpower,but also is difficult to meet the real-time and accurate analysis.The vehicle detection and tracking system based on deep learning can achieve higher accuracy,higher real-time and more intelligent automatic analysis.As the focus of highway traffic operation management,vehicle detection and tracking is of great significance.Therefore,this paper studies and implements a set of road traffic safety vehicle detection and tracking system based on deep learning.The main research work of this paper is as follows:Firstly,the theoretical knowledge of vehicle detection and tracking algorithm is introduced,which lays a solid foundation for the subsequent abnormal behavior analysis.Then,the working principle of the mainstream algorithm in the current target detection is analyzed.From their detection effect and performance,the detection accuracy of yolov3 and yolov4 algorithm meet the actual requirements,but the detection speed is only about 2-5 frames(RAM is 16 g The processor is Intel(R)core(TM)i5-10210 u @1.60 ghz 2.11 GHz,which can not meet the real-time requirements.Although the detection speed of yolov3 tiny algorithm meets the real-time requirements,there are some defects in the detection accuracy,such as missing and false detection.But the error of detection accuracy of yolov4 tin algorithm is within the error range,and its detection speed is also the fastest.FPs in my notebook has reached about 40 F / s,which fully meets the real-time requirements.Therefore,yolov4 tin algorithm is used as the basic algorithm of vehicle detection in this paper.Then according to the analysis of the target detection algorithm and the current multi-target tracking algorithm,this paper will complete the road multi vehicle tracking task based on the "detection and tracking" deep sort algorithm.Firstly,the Kalman filter is used to predict the bounding box parameters of each detected target,and the Kalman filter is selected through the experiment After that,the Hungarian correlation algorithm is used to measure the IOU values of detection results and tracking results to get the correlation results between frames.Finally,based on the above vehicle detection and vehicle tracking task research,this paper builds a vehicle detection and tracking management system in highway traffic,and on the basis of tracking,the target trajectory points are obtained for clustering analysis.The whole system includes front-end display module and background processing module.The system separates business processing and data display logically,and visually shows the effect of vehicle detection and tracking in the front-end page,and processes the data in the background.