Tracking Ball and Players with Applications to Highlight Ranking of Broadcasting Table Tennis Video

This paper presents an algorithm of automatically ranking the highlight levers for broadcasting table tennis games. We rank the highlight lever according to some extracted high-lever semantic features, including the table position, the player action and the ball trajectory. The table is located by the edge and color features in a rough-fine procedure, while the player is tracked by the changing mask and a trajectory analysis. To derive the ball trajectory, all ball candidates are first detected using color, shape, size and position limits. Then both the motion and appearance information are employed to track the ball by a Bayesian decision framework. The ball's dynamic and appearance parameters are updated by a Kalman filter and an incremental Bayesian algorithm. With those semantic features, we rank the basic highlight lever with the feature statistic and measure the quality of the game from a fuzzy system.