Research On Detection Method Of Vehicle Abnormal Behavior

In recent years,with the continuous improvement of our economic level,the average consumption level is increasing increasingly,more and more people choose motor vehicles as transportation tools.The number of motor vehicles in China ranks among the top in the world all year round.With the surge in the number of motor vehicles,the frequency of traffic accidents is increasing year by year,which seriously threatens the interests of the country and the people.At the present stage,the main role of urban road surveillance video lies in the evidence of traffic accidents and the review of events for the traffic supervision department.At present,the urban road surveillance video system is in a semi-automatic state,which mainly relies on the personnel of the traffic supervision department to manually query and locate the place of the accident and the vehicles involved in the accident.This kind of method can only realize the post-incident review of traffic accidents,but cannot prevent the occurrence of traffic accidents so as to reduce the incidence of traffic accidents.Moreover,this method has extremely low detection efficiency and contains serious errors.This topic is based on the traffic surveillance video object detection,object tracking,and target abnormal behavior detection three links to carry out in-depth research,and improve the algorithm within each link and select and develop a detailed research route.(1)Object detection: In order to achieve the research objective of this topic to achieve real-time detection of vehicle abnormal behavior,the vehicle target should be detected from the surveillance video first.In order to improve the stability of the background difference algorithm,the Surendra background update algorithm is adopted on the basis of the original algorithm structure to reduce the complexity of the calculation,improve the detection efficiency and realize the threshold adaptive,and the three-frame difference method is added to improve the stability of the original algorithm.The experimental results show that the algorithm has high detection efficiency and strong anti-interference ability,especially in solving the problem of light mutation in the external environment,and its detection results are real-time.(2)Object tracking: On the basis of object detection,the center of gravity of the sideways target and the length and width of its outer rectangle are defined as global features and extracted,and the similarity function is proposed for feature fusion.By constructing the interval of interest and tracking list,this topic can effectively solve the multi-target tracking and reduce the interference of targets except vehicles to the tracking results,so as to realize the realtime and stable tracking of the measured target and extract its trajectory.(3)Abnormal behavior detection: On the basis of object detection and object tracking,this topic develops detailed detection principles for the three most harmful vehicle abnormal behaviors of speeding,reverse traffic and red light running respectively.Meanwhile,multiple sample videos are selected for multiple groups of comparison experiments,and the detection results are verified.Finally,the experimental results prove that the vehicle one-field detection method in this research has strong stability compared with similar methods,and the detection results are both accurate and real-time,which is in line with the research objectives specified in this project.