Sports video analysis

Abstract

Summary form only given. Techniques of computer and Internet are developing very fast, and users want to access their interested multimedia information from anywhere at anytime by using their most convenient digital equipments. Sports video always appeals to large audiences, and it becomes an important problem to automatically extracting useful semantic information from sports video to facilitate user's accessing requirements. Therefore, sports video analysis has emerged as a hot research point in multimedia area recently due to its high commercial potentials. This talk gives an introduction on technologies and prototypes of sports video analysis and enhancement that include object detection and tracking, scene analysis, event detection and highlight summarization, as well as 3D virtual sports event reconstruction. In the processing of sports video, playfield, player, ball, overlay text and jersey number are detected in the object level. Video scenes with the same semantic meaning are categorized together by using supervised classification and unsupervised clustering algorithms. Solutions of event detection and highlight ranking by using multi-model information are employed in the highlight extraction level, which is one of the most important goals in sports video analysis. 3D virtual cartoon from broadcast soccer video are generated by applying both computer vision and computer graphics techniques. These results can be used to intelligently adapt, enhance and personalize the multimedia content to meet users' preferences and network/device environments.