A tennis video indexing approach through pattern discovery in interactive process

Tennis game is a mutual and interactive process between the two opponents, and audiences are usually interested in how the players oppose and compete with each other. However, previous works on tennis video indexing are mainly based on analyses of player's behavior, such as movement and gesture, and seldom provide information of the match process. In this paper, a novel approach is proposed for tennis video indexing by mining the salient patterns in the match process. The match is first characterized by a time sequence of players' joint movements, based on which an unsupervised way is performed to extract the sub-sequences with high occurrence frequencies. These sub-sequences, called patterns, reflect the technical styles and tactics of the players, and offer an effective manner to organize the video. Evaluations on four hour broadcasting tennis videos show very promising results.