A Survey on Event Recognition and Summarization in Football Videos

There are large number of sports videos are available over the World Wide Web (www). But using these complete videos to get information is a hectic and time consuming job. So, the need of extraction on events from the videos aroused. Semantic analysis of videos and automatic event extraction plays a vital role in several applications; using content-based search engines, video indexing, and video summarization. As a powerful tool for learning complex patterns is the Bayesian network, this paper proposes a novel Bayesian network (BN) based method automatic event recognition and summarization in Football videos. This method includes efficient shot boundary detection algorithms, shot view classification, and the related Bayesian network construction. There are three main stages: The shot boundaries are detected in first stage. Using some model, the video is segmented into the play-break sequences, large but meaningful semantic units. Within the next stage, theses play-break sequences are used to extract several key events. Lastly, in the final stage, the Bayesian network is used to get the high level semantic events. Constructing the Bayesian network is the basic part of the method. By applying family of Copula, the joint distributions of arbitrary variables of the network are modeled. Some of events that can be recognized by this method in Football videos are goal, card, corner, shots on goal, foul, offside, missed shots and non highlights. The users are more likely to be interested in these events and not in complete and large videos.