Unified Sports Video Highlight Detection Based on Multi-feature Fusion

This paper presents a novel method for highlight extraction in sports video, which fuses multi-modal features including audio and visual. At first, we detect slow-motion replay based on logo, and then detect audience cheers based on pitch frequency and short-time energy. In the next step, we fuse audio-visual features to detect highlight events through using a special "and" (otimes) operation. The method is general to most genres of sports game because the audience cheers and slow-motion replay are a reliable clue to the highlight and the features we utilize is not limited in a specific kind of sports game. Experimental results show that the approach is valid and avoid the deficiency of detecting highlight events by a single feature.