Detection And Classification Of Non-motor Vehicle In Video Sequence

Video GIS is a geographical environment perception and analysis platform which consists of video analysis system and GIS, it can manage, analysis and specialize the video data, reconstruct the video geographic scene, integrate the reality and the virtual scene and display the fusion of it in a unified geo-referenced model. Mixed traffic is a common phenomenon in China’s urban transport, but Intelligent Transportation (ITS) technology focus more on extracting the information of motor vehicles, there is few studies focus on accessing the information of non-motorized vehicles. This article studies on the classification of non-motor vehicles that in the road traffic video monitoring. This study focuses on the image transformation method, the detection of moving target in traffic surveillance video, the tracking and the classification algorithm. The main research work and results are as follows:First, the classification of non-motor vehicles and the characteristics of each class were defined. Based on the actual situation of the urban road traffic, some class of non-motor vehicles is added to make a supplement of the non-motor vehicles’ classification which was provided by Road Traffic Safety Law of the People’s Republic of China. According to their morphological features in a video sequence, its further detailed classification was defined, and summed up the shape and motion characteristics of the various types of non-motor vehicle.Second, overall methods of non-motor vehicle classification was established. Gaussian mixture background subtraction method was selected to extract moving foreground target based on the image processing technology, so the background model can be build fast and stable, and the foreground objects can be accurate and stable. Target tracking chain structure has been designed with its generating and demising algorithm. Matching characteristics of the target has been used to finish the target tracking. Image perspective transformation technology has been used, so that each pixel in the surveillance video can get a real location, which can get moving target’s movement parameters. Classification of the non-motor vehicle has been accomplished by support vector machine that using features such as the average offset and maximum offset of the moving target’s trajectory, the speed and the deformable part model of the moving target. For the swarm case of non-motorized vehicles on the road, area threshold method has been used to estimate the specific number.Third, the non-motor vehicle inspection and classification system has been established by using the traffic video surveillance. The core of the system consists of moving target detection, moving target tracking, classification and traffic parameter extraction. The experimental verification system through the proposed method, the effect is good.