Research On Target Detection And Classification In Video Stream

As the rapid development of the network, communication and microelectronictechnology，a number of vision analysis system with specific function get more andmore attention for its intuitive, convenient and rich in content. Traffic monitoring is themost widely used area. However, the all-weather surveillance system captured a largenumber of video information, in which it is inefficient and mistakable to find targets viapure-manual search methods. Therefore, it was hoped that computers can analyze andunderstand video content, in order to achieve intelligentized and practical video analysissystem, video analysis techniques have emerged.Video analysis techniques deal primarily with video sequences which contain avariety of moving target, and detect, track, classify and recognize targets, thenunderstand and describe their behavior. Target classification is an important aspect ofvideo analysis with the content of classifying the object area based on motion analysisby their features of shape and motion, and is important to the development of high levelvideo understanding techniques.In this paper, we make a research on methods of moving target classification basedon static camera, and realize the classification between pedestrians and cars in videostreams. In this paper we propose a novel shape character：Normalized View DirectionDistance. With this new character and some common shape characters, through a systembased on background modeling and Support Vector Machine (SVM), each target of eachframe is classified. The classification results make it out that our method can effectivelyidentify pedestrians and vehicles in a video stream, and is robust to background shelterand scramble backgrounds.