Soccer video semantic concept detection based on Bayesian belief network approach

In this paper a method for detecting semantic concepts in soccer video based on Bayesian Belief Network (BBN) classifier is proposed. In most broadcast soccer videos, replays succeed excitement clips. Replays often stand between two successive logos. Here neural network learning is used to detect logos. Events such as close-ups of players, close-ups of referees, staple line in corner point, spectators and players' gathering are extracted from the excitement clips. These events are used as evidences for BBN and posterior probabilities of semantic concepts such as goals, saves, off-side, foul and corner are computed. Experimental results on several soccer videos demonstrate the effectiveness of the proposed approach.