Algorithm Research On Automatic Identification Of Traffic Accident Based On Video

From the beginning of twenty-first century, Chinese society and economy hasobtained fast development, and urbanization process has got a constantly acceleration.Besides, traffic congestion in urban is getting more and more serious, which has led toan increasing pressure of urban road traffic. What is worse, the expanding numbers ofvehicles have caused more traffic accident on urban roads. Traditional traffic accidentdetection and inspection are running in manual monitoring methods. But thesemethods have low efficiency and poor real-time. With the rapid development ofintelligent transportation system and the wide application of computer visiontechnology, using video image processing technology for real-time analysis of roadsurveillance video, intelligent detection accident, and getting traffic information hasbecome a research hot spot. Real-time detection of traffic accidents could not onlyreduce the waste of police resources, but also have great significance for improvingthe efficiency of dealing with traffic accidents.This paper studies on the automatic identification algorithm of traffic accidentbased on video method. The main work is as follows. Firstly, this paper introduced thetechnical difficulties of target detection and summed up the common methods ofinitializing the background model; then used average background modeling method toextract the background image, and after, updated the background to current statethrough the background update algorithm. Secondly, extracted difference image usingbackground difference method, and calibrated the connected area after the vehicleshadow removing, realized the moving target detection. Through the experiment andits results, the paper verified the effectiveness of this algorithm. In the part of targettracking algorithm, the Kalman filter method was proposed and applied to carry outthe moving target tracking and matching, using the centroid distance and area size ofthe moving foreground as the matching parameters. Again, according to the movingtarget detection and tracking results of the previous chapter, extract the motioninformation of the foreground object; calculate the moving target speed, acceleration,direction of travel of the radian value and other features. Using Kalman filter topredict the moving target centroid point coordinates of the next frame, and comparewith the detected centroid position of the foreground, and determines whether thedetected foreground is coincidence or not, if two foreground objects overlap occurs, the traffic conflict is happening. Then, make the further judgement for traffic conflicts,the algorithm of identifying traffic accident automatically is proposed, whichsynthesize vehicle deceleration, driving direction change rate and time parameters,and identified the occlusion caused by vehicle and the phenomenon of pseudocollision, ultimately determine whether the traffic accident occurs. Finally, theexperiment was verified. In addition, the discriminant algorithm was proved thevalidity. At the same time, the paper analyzed the error of the experiment.