Real-time visual play-break detection in sport events using a context descriptor

摘要: The detection of play and break segments in team sports is an essential step towards the automation live game capture broadcast. This paper presents a two-stage hierarchical method for play-break non-edited video feeds sport events. Unlike most existing methods, this algorithm performs action event recognition on content, thus does not rely production cues broadcast feeds. Moreover, require player tracking, can be used real-time, easily adapted to different sports. In first stage, bag-of-words detectors are trained recognize key events such as line changes, face-offs preliminary play-breaks. second output along with novel feature based spatio-temporal interest points create context descriptor final decision. Experiments demonstrate efficiency proposed real hockey footage, achieving 90% accuracy.