Research On The Theories And Methods Of Detection And Recognition For Traffic Violations Based On Video Technique

In recent years, traffic safety in our country has become a big anxiety. Asnational economy develops and people's living conditions improves, the numberof people owning cars is on the sharp rise, as a result accidents and deaths are alsoon the sharp rise. According to statistics, more than 100,000 people died in trafficaccidents in the year 2005, which amounts to a cruel battle. Though the causes ofthe accidents are different, most of them are closely related to violations ofpedestrians or vehicles. Therefore how to find and correct traffic violations is amajor issue confronting every traffic participant and supervisor. With fastdevelopment of computer science, information processing, video technique andartificial intelligence, an epoch-making opportunity has come to employtechniques in coping with traffic violations.Referring to the science and technology development project of Jilin provinceTraffic Violations Processing System Based on Pattern Recognition, the importantproject of Heilongjiang province Freeway Traffic Monitoring and AnalyzingSystem by Video Technique and the sub-project "Study of Fast ProcessingTechniques of Traffic" of the important project sponsored by National NaturalScience Fund Study of Key Theories and Simulation Techniques of Dynamic TrafficManagement and Control of Urban Roads, and in combination with my recentresearches made in my participating in projects and the relevant literature bothhome and abroad, the thesis makes a deep and extensive study of the theories andmethods of detection and recognition of traffic violations based on video technique.The contents are as follows:(1) Pretreatment of Video Image of Traffic sceneDue to conditions of illumination, weather, equipment and the like, videoimages of traffic scene may not be desirable, and therefore the video images needto be pretreated, so that they can meet the requirements of real time and accuracyof detection and recognition of traffic violations. Based on the analysis andcomparison of various types of commonly used filtering, various multi-contrastreinforcing algorithms and various edge detection algorithms, the pretreatment oftraffic scene images by employing median filtering, reinforced template andmorphologic edge detection algorithm is proposed, and the algorithms are provedeffective through experiment.(2) Theories of Classification of Traffic Detectors and Video VehicleDetectionsTheories of classification of traffic detectors are introduced, while principles,performances and features of various types of traffic detectors are analyzed andsummarized;micro methods of traffic video detection are discussed, such asbackground difference, interframe difference, interframe corresponding and edgedetection, etc. The algorithms of vehicle detection based on morphologic edgedetection and background difference are proposed and is proved effective throughexperiment.(3) Key Techniques of Vehicle Video DetectionKey techniques that affect vehicle video detection are dealt with in length,which involve background refreshing, threshold selection, shadow processing, andseparation of hidden vehicles. This part of paper mainly deals with selectivebackground refreshing algorithm based on background adjustment, histogram forthreshold selection of gray scale, threshold selection of color image and selectionof dynamic self-adaptive threshold, analysis of vehicle shadow feature andpartitioning detection of shadow. For vehicles slighted hidden, simple separativealgorithm is given.(4) Detection and Recognition of Vehicle Violations in the Traffic SceneSpecific detection and Recognition algorithms are offered to deal with thefour common violations of cutting across red light, wrong parking place, speeding,making a wrong turn. The questions that may occur in the algorithms are dwelt onand their solutions are given.(5) Saving Images of Violating Vehicles and Database RetrieveA study of compression algorithm of vehicle image and retrieval algorithm ofvehicle image feature is made. After sufficient study of violating vehicleinformation, a violating vehicle database retrieve method is acquired based onvehicle color, violating site, violation type, and violating time. With all-roundviolation information, efficient violation supervision is achieved.(6) Design and Realization of the Processing System for Violating VehiclesFor the four common violations (cutting across red light, wrong parking place,speeding, illegal right-hand bend), in the environment of Windows Me and makinguse of the daughter environment of VC++6.0, an integrative monitoring andsupervision system of traffic violations including video data picking from thetraffic scene, violations monitoring, violating vehicle's panoramic image savingand management of database of violating vehicles is achieved with the emphasison the design of system hardware and software and principles of realization ofsystem functioning.At the end of the paper, a conclusion is made of all work with an analysis ofits improvement and achievements, and further research is presented.