A Research On The Retrieval Of Digital Video Based On Motion Features

With the rapid development of multimedia computing technique, the traditional retrieval methods based on keywords & description texts are not competent for people. The demand for effectively managing abundant multimedia resource becomes urgent day by day. Content based video retrieval technique arises at the historic moment. It means to search and retrieval in the extensive multimedia database according to media content and the context of target. CBVR has been paid more and more attentions and shall be the method to solve the problems as mentioned above. This paper has done the following work around this field:1. The structure of digital video is complicated and different from traditional data based on characters. It is essential for video retrieval to set up a rational video data structure model. On the basis of analyzing the existing video data model, we put forward a kind of comparatively suitable data model for video with the help of MPEG-7 standard, and describe its semantic features by XML.2. Shot change detection is a foundation work in video analysis. In this paper, we study most existing algorithms, and point out their merits and shortcomings. According to the study of video spatio-temporal slices, we propose a novel and fast algorithm for video shot abrupt change detection.3. Motion features in video can usually be divided into two types. They are global motion, which is caused by camera motion, and local motion, which is caused by video object motion. According to the study of single-shot spatio-temporal slices, we propose an algorithm for camera motion detection in this paper. When camera is static, we first get moving change region by difference analysis, then detect moving object through combining moving change region and edge information in the current frame, then gel its motion information.4. At last, we design a real-time CBVR system through video clip with the algorithms which are proposed before.