Automatic Sports Video Analysis using Audio Clues and Context Knowledge

Sports analysis has recently become popular in research and professional applications. This paper presents a scheme for automatic sports video analysis based on audio clues and specific game context knowledge. We propose a simple, two-step racket-hit detection for achieving accurate event classification for tennis video. To implement the mapping between the sample-level feature space and the semantic-level space, we employ heuristic rules based on specific knowledge of the tennis game. Experimental results have shown that the proposed system can reliably detect the racket hit (at about 90%) and identify meaningful events such as rally, scoring, different types of service, and return. Our system can be operated stand-alone or combined with video analysis and then used for effective and automatic extraction of various tennis events and analysis of tactics with high reliability.