The Study On The Abnormal Events Detection Algorithm Of The Urban Traffic Based On The Video Surveillance

The detection of traffic abnormal events based on the video surveillance processinghas become one of the vital application technologies. Compared with the traditionaltechnology of buried coil detector for the detection of traffic incident, this technology hasbetter real-time and intuitive. However, some key technologies about detection system ofabnormal events are still on the development stage, especially in traffic behavioral analysisof moving target and tracking for the traffic events, it is need to be studied further.In this paper, the research object is the surveillance video of traffic crossings in thecity, some key technologies of traffic abnormal events detection in the city trafficsurveillance video are researched systematically by using graphic processing knowledge,pattern recognition and machine learning. First of all, the establishment, extraction andupdating of background models are fully discussed, the statistical histogram backgroundextraction method based on the IPM vision transformation is proposed which overcomesthe problem that vision image is mistakenly counted as background; The backgroundadjustment is applied to update the background, and the solution for removing the motionshadows is raised at the same time which can increase the accuracy of moving targetdetection. The scale space Mean-Shift multi-targets tracking method with joint ofmulti-features is structured by combining the colors of moving targets and HOG features,which overcomes the bad effects of single feature tracking and changes of trackingperformance caused by the target places in the background; Then estimate the motion stateby Kalman Filtering, and thus, the overlap issues caused by the contact of moving target,the robustness and precision are increased when it comes to the multi-targets tracking andlocus extraction. Considering the space features and direction features of moving targets,the city traffic abnormal events detection method based the mixed patterns matching isproposed in this paper, the abnormal events will be detected by the mixed pattern matchingon space and direction when the moving targets proceed into the detection area. The method in this paper is tested and validated by recognizing the lane changing events andopposite moving against the direction of lanes in actual city traffic scenes.The theories and methods of traffic events video detection is deepened by the researchachievements in this paper, and an effective technology on the traffic behavioral features ofmoving targets and information collection of smart traffic events are offered by theresearch which has important theoretical significance and practice value.