Research And Application Of Vehicle Detection And Tracking Algorithm

In recent years,the potential of China's auto market has continued to be released,making China's traffic environment worse.The traffic order is chaotic,and traffic accidents are frequently becoming more serious.The continued increase in car holdings has put a huge test on intelligent traffic monitoring systems that are still in the research and improvement phase.Most of the traditional intelligent traffic monitoring systems only have the ability to record video data without the ability to analyze and understand video object detection and video event behavior.For this kind of traditional video surveillance system,there are some shortcomings:through manual observation and observation,it takes a lot of human and financial resources,and the efficiency is low;it is impossible to observe the position of each abnormal situation,and the judgment is inaccurate and missed.Situation;poor real-time,unable to predict unexpected events.Therefore,research on vehicle detection,tracking and abnormal behavior recognition based on surveillance video image processing is of great significance for improving traffic management methods and ensuring road traffic safety.In this paper,from the perspective of practical application,the surveillance video sequence is taken as the research object,and several key technologies such as lane line detection,vehicle target detection,multi-vehicle target tracking and abnormal behavior recognition are studied.Combined with vehicle detection,vehicle tracking results,and road information of lane lines,three automatic behavior detection algorithms for parking,retrograde and illegal lane change are designed.The specific research mainly includes the following aspects:(1)A preprocessing method based on the combination of Sobel edge feature and HSV color feature is designed.The position of the lane line is roughly located,and then the Hough transform is used to extract the current coarsely positioned lane line to determine the lane line position.(2)A vehicle detection method based on YOLOv3 is designed.In the aspect of vehicle detection,the traditional algorithm and the deep learning-based method are used to compare the experiments.Finally,the Yolov3 algorithm with relatively accurate detection results is selected to realize the vehicle detection,and the detection result is used as the follow-up multi-vehicle detection-based tracking(Tracking-by-Detection,TBD)input response of the framework.(3)A multi-vehicle target tracking method based on Markov Decision Process(MDP)is designed.Using the TBD tracking framework,the vehicle target detection result is input excitation,and a decision-based Markov process is adopted for each vehicle target to establish an MDP model to achieve accurate tracking of multi-vehicle targets.(4)According to the results of vehicle detection and tracking,the multi-vehicle motion trajectory is drawn,combined with the trajectory parameter characteristics to model and analyze,and the discriminant models of three abnormal behavior events,such as parking,retrograde and illegal lane change,are established.The experimental results show that the selected algorithm can accurately detect and track the vehicles in the surveillance video,and combine the driving trajectory and lane line to effectively discriminate the abnormal behavior of vehicles in the video.