Study On Background Modeling And State Identification Method For Highway Congestion Event Detection

Due to the sealing of the scene and the high driving speed, the highway trafficcongestion Influences traffic capacity, it will easy to cause secondary accidentsresulting in casualties and severe economic losses. Video surveillance is an essentialmeans of the highway monitoring, a lot of video surveillance equipment has installedat the key sections of highway. For the highway scene characteristics of complexityscene and video impacted by environmental easily traditional congestion identificationalgorithm can’t get satisfied result in highway, so study of highway traffic congestionstate identification is more and more important to improve highway capacity andoperational safety.Vehicle target segmentation and traffic parameter selection is the basis of thetraffic congestion state identification, existing video-based traffic congestion stateidentification for vehicle target segmentation, traffic parameters selection at targetincomplete and traffic congestion state identification error detection reducing undercomplex environment is less study. Based on analysis of existing traffic congestionidentification algorithm, this paper focuses on vehicle target and traffic congestionstatus identification, including background modeling, traffic parameters selection andtraffic congestion state identification error detection reducing，eventually formed a setof highway environment traffic congestion identification method based on video.In the part of vehicle target extraction, for the highway scene characteristics ofcomplexity scene and video impacted by environmental easily, non-parametric kerneldensity estimation method is used to get background, and corresponding backgroundupdating method is given in both cases of scene light changing gradually and suddenly,when traffic flow is large, background can’t be accurately established, so an initialtraffic flow algorithm based on fractal dimension is given, background subtraction andmorphological methods are used to do vehicle target segmentation. Experimentalresults show that non-parametric kernel density estimation method can get betterbackground, background updating method can adapt to the abruptly and graduallychanges of background, initial traffic flow algorithm based on fractal dimension canjudge the size of initial traffic flow, background subtraction and morphologicalmethods can get vehicle target accurately.In the part of traffic congestion status identification, a method based on fuzzy C-means is proposed. Firstly for incomplete vehicles target detection causing sometraffic parameters get inaccurate, through the analysis of common traffic parameters,average space occupation and time occupation are selected, then fuzzy C-meansclustering algorithm are used to obtain cluster center of each status from sample data,finally using Euclidean distance to determine congestion state of current traffic. Toreduce the false detection of congestion state, an algorithm based on connectedcomponent analyzing and vote are given to eliminate false detection and improve theaccuracy of traffic congestion status identification.Finally, combining of these algorithms, highway traffic congestion statusidentification system is build, using multiple Chongqing highway scenes surveillancevideo and in VC environment to do experiments. The results show that this method canachieve vehicle target accurately and get the status of current highway trafficcongestion fast and accurate.