Sports Video Analysis

Shot Classification

In a sports game, the positions of cameras are fixed around the field and each camera has a specific assignment for broadcasting the game. The rules of presenting the game progress are similar in different channels. The broadcasting technique that a few dominant shots constitute most parts of a sports game leads to the requirement of shot classification. Duan et al. (2003 & 2005) employ a supervised learning scheme to perform a top-down shot classification based on mid-level representations, including motion vector field model, color tracking model and shot pace model. Hua et al. (2002) integrate color distribution, edge distribution, camera motion, sound effects and closed captions with maximum entropy scheme to classify baseball scenes. Kumano et al. (2005) divide a frame into blocks and analyze the mean, variance and log variance of the luminosity within each block for pitch scene discrimination. Lu and Tan (2003) propose a recursive peer-group filtering scheme to identify prototypical shots for each dominant scene, and examine time coverage of these prototypical shots to decide the number of dominant scenes for each sports video. Mochizuki et al. (2005) provide a baseball indexing method based on patternizing baseball scenes using a set of rectangles with some image features and a motion vector.