On Video-Based Vehicle Violation Detection

With domestic economy’s high speed development, accelerating urbanization process and increase of the pressure for urban traffic regulation in recent years, the effects of all kinds of traffic violation behavior is going from bad to worse. As a subsystem of the intelligent transportation system, the video based Vehicle violation detection provides the basis for the traffic administrative department of traffic regulation and decision through the study of the automatic detection of vehicle traffic violations and records. This paper combined with the present video vehicle violation detection research and do the research from the following several aspects.First of all, the moving target detection was being studied. By comparative and analysis of several major foreground detection algorithm such as adaptive background updating, Gaussian mixture model, fuzzy mathematics and neural network, the ViBe background extraction method is chosen to achieve the goal of foreground detection, at the same time, because of the shadow caused by the foreground detection are widespread, Results in the decrease of test results, which affects the subsequent illegal behavior criterion, In this paper, the shadow detection method based on HSV color space characteristics is used to the ViBe algorithm ,as much as possible to ensure algorithm processing speed at the same time realize the shadow removal and inhibition.Secondly, Vehicle tracking and trace extraction were studied. In the process of vehicles tracking, In order to achieve the discriminant of peccant vehicles need to extract information of vehicle target track, based on the characteristics (center of mass, high prospect target connected domain wide and unit time migration) extracted during the Foreground detection process, on the basis of through the multiple correlation between feature matching method to realize the vehicle, and to establish the stable tracking multiple vehicle tracking list. At the same time in order to improve the searching efficiency of the algorithm, using kalman filter algorithm to estimate the vehicle state information vector.Thirdly, detection of vehicle traffic violations was being studied. We put forward specific test menthol for vehicles peccancy retrograde, illegally turning, Violation of line pressing and illegally parking. By compared the related transportation information such as the trajectory, the starting frame and the starting position saved by the tracking target list with the actual road traffic scene for detection of vehicle traffic violations.Finally based on the existing different algorithms libraries, this paper integrate all the algorithms that vehicle detection research involves, and develop a test system platform integrated 35 kings of foreground detection algorithm,5 kinds of shadow removal algorithm and 5 tracking algorithm, to provide convenience for follow-up studies.