Tennis Video Multi-level Analysis And Retrieval

Sports video is a kind of important multimedia data and has attracted people all over the world. People are not satisfied with just watching match or browsing sports video. There are diversified demands, such as the wonderful clips summary and the particular event detection, etc. These services are dependent on the semantic analysis and understanding of sports video.In this paper, tennis video is considered as research objects and analyzed in three levels, including feature-level, object-level and semantic-level. The main research contents and results of this paper can be concluded as follows:(1) Based on analyzing color characteristics of the tennis court and shape characteristics of the court-line, an algorithm, combined domain color with shape characteristics of the court-line, is presented to detect tennis court clips. More accurate results can be obtained.(2) Based on former researches, a new algorithm, which based on the adaptive GMM, is exploited to detect court line. This method can be used to extract tennis court-lines and reconstruct the complete tennis court-lines.(3) The improved CamShift(Continuously Adaptive Meanshift) algorithm is used to track athletes, which can learn more about the semantic relationship of object targets in space and time, and semantic-level events can be distinguished.(4) Combined with the analysis of object-level and semantic-level, features of tennis matches are applied to construct a linear model, which can used to detect and identify semantic-level events in tennis match.According to the above ideas and algorithms, a prototype system for content-based tennis video analysis is designed by using Visual C + + 6.0 and the Intel OpenCV Library. Experimental results indicate that analysis algorithms for all levels are more effectively.