Content-based Table Tennis Games Highlight Detection Utilizing Audiovisual Clues

Abstract

Both audio and video are considered as important information carriers of multimedia content. In this paper, we propose an algorithm utilizing audiovisual clues for a scenario of sports game highlight detection, where the highlight detection for table tennis games are studied. Since audio and video contain different aspects of information that is helpful to locate highlights, we build two algorithms detecting highlight candidates based on audio and video, respectively, where hidden Markov model (HMM) audio keyword modeling and unsupervised shot clustering are applied. Decision fusion is invoked to combine audio and video highlight candidates and generate final highlights. Promising experimental results up to 90 % average precision are achieved.