Reaserch On Automatic Discrimination Of Traffic Conflict Based On Video

With china entering the automotive era, traffic accidents also increase up, resulting inhuge economic losses. Therefore, effective on intersections or sections of traffic safetyevaluation has become an important research topic. Because after the onset of trafficaccidents, traffic safety evaluation by accident has some drawbacks. So, based on trafficconflict traffic safety evaluation methods have been proposed.Although the traffic conflict approach can better evaluate traffic safety, however, thecurrent method of collecting traffic conflict mainly through artificial means of observation.Thereby, it can have some measure of the inconvenience. So we presents a video-basedtraffic conflict automatic identification technology. This technology can automaticallydetect the traffic conflicts, solve the inconvenience of traffic conflict on intersection orsection.But the current video-based automatic identification system for traffic conflicts existmany problems, such as imprecise of target detection, tracking ineffective, conflictidentification accuracy rate low. So, based on the lack of current research, our article makea thorough research. The main work is as follows:(1) Target detection algorithm based on background subtractionTarget detection is the first step of the system, in order to solve this problem,comparing the advantage and disadvantages of each target detection algorithm, we use thebackground subtraction algorithm to extract foreground object. The algorithm first extractsbackground image using background model and update the background image using thebackground updating model. Then we use the background subtraction algorithm to getbinary foreground image, and use the various regional connectivity calibration algorithm toget foreground object. Finally, we use the target classification algorithm to obtain the targetclassification. Through experimental analysis, the algorithm achieved a good detectionresults.(2) Target tracking algorithm based on online learning Then the after step is target tracking, this paper proposes an improved online boostingobject tracking algorithm. Since the original online boosting tracking algorithm exist theproblem of real-time performance pool and drift to left turn vehicle, this paper improvesthe algorithm. Firstly, this paper presents a cascade classifier to improve tracking speed.Then, we propose a main direction model to improve tracking performance, solving theproblem of existing tracking drift. Finally, we propose the target position prediction modelto reduce the search area, and speed tracking. Through experimental analysis, thisenhancement algorithm is better than traditional online boosting tracking algorithm inspeed and accuracy.(3) Traffic conflict automatic identification algorithmAfter acquiring the target detection area of all moving objects in real time andhistorical trajectory and velocity data and other parameters, it is necessary for real-timeautomatic traffic conflict identification. First we establish a traffic conflict discriminationmodel based on critical distance, it is able to discriminate against traffic conflict. Then,based on video processing, we establish the traffic conflict discrimination processes.Comparing the tradition method and my method, from the result of experiment, our methodis fast and accurate.This paper presents an algorithm of automatic traffic conflict discrimination, it deepenresearch in this field, and give service for intersection and road safety evaluation. It hasimportant practical value.