Sports Metadata Extraction Technology Research And Implementation Based On Video Content

With the development of computer technology, multimedia and significant increase of information, the information needs of people become more and more popular. Multimedia information, which due to intuitive and content-rich, is loved by the masses and become the main sources of data. In recent years, interactive video applications led to the boom of content-based video analysis, people turned to the needs of multimedia content-based access, retrieve and manipulate the characteristics of interactive video services, video applications and a variety of media, entertainment video-on-demand. Video information, because of its unstructured data characteristics, can not be directly used in keyword searching. The work of manual annotation is very large and has quite subjective and arbitrary. If the video own a structured meta-data index based on the video content, it can be managed efficiently.Since MPEG-4, MPEG-7 and other standards were proposed, the understanding and extracting information for video retrieval became a very important area for research. Although the use of regional characteristics of image segmentation research has made great achievements, extraction of semantic objects and establishment of robust multi-level semantic information are still urgent issues. The main tasks of this paper are learning video segmentation and content analysis technologies, establishing a general, scalable physical sports video metadata description with knowledge of sports areas.The main idea of this paper:on the basis of previous research about sports video semantic description model, a multi-level video content-based metadata description model and related algorithms were proposed. Specific process as follows:first, after shot detection, video sequence was divided into a set of shots, then extracted key frame from a shot. With video segmentation algorithm, distinguished objects in a frame. Recognized the same object in adjacent frames using feature space clustering, tracked object movement. Finally, a sport video metadata extraction system was realized based on the above analysis.