Sports Video Retrieval Research

Sports videos, which have a stake in modem people's life, are getting popular day byday. With the growing amount of sports video, it has become an emergency to help usersfinding out their favorite games and highlight. This paper takes tennis and badminton videoas research objects, and discusses several problems in the process of content-based videoretrieval, aiming to generate sports video summary for users using computational processand analysis.In this paper, several techniques are discussed including shot boundary detection, keyframe selection, feature extraction and shot classification.Shot boundary detection is the first step of video processing, starting from ageneralization of methods of shot boundary detection in compressed domain andin-compressed domain, this paper proposed an improved spatio-temporal slices-based shotdetection algorithm in in-compressed domain and an approach using macroblock type incompressed domain. The second step of video processing is key frame selection and shotclassification, in this part, key frames and their features, which lay a good foundation forshot classification, are extracted using some low-level features such as color and texture,followed by cluster processing based on ball tree model. To complete the classification ofbadminton and tennis, this paper divides the court-lines from the key frames usingdominant color detection, robust cylindrical metric and morphologic method, and do theclassification by using the court-line's character.Finally, this paper implements these algorithms by Matlab, and the experiments havedemonstrate that all these methods are effective.