More Clues To The Fusion Of Football Video Semantic Analysis And Event Detection

In this paper, soccer video is taken as the research object, in which the typical event detection skill is analyzed and researched. Cues are introduced as an element between the low-level features and the high-level semantics in content-based retrievals. From soccer video characteristics, combined with the soccer play knowledge, cues are extracted out from low-level feature and integrated to detect the shooting event, foulness event and general event in soccer video based on the hierarchical hidden Markov model (HHMM).Shot classification is an important research content in soccer video processing and retrieval. Considering the shortcomings of the existed methods, we propose a novel approach to classified shots into four types:long-view shot, middle-view shot, close-up shot and out of field shot using location essential color ratio, figure number and average area and so on.The slow motion is closely related to the exciting events of the football video. A method for slow motion detection based on logo characteristics is proposed and it can determine whether a given shot consists of a slow motion.The input video stream is first segmented into shot sequences, then, shot classification, slow-motion detection, goal detection, caption detection and audio classification are extracted as semantic cues from the shots based on low-level features, and, HMM models of shooting event, foulness event and general event are built and trained to infer the events from the cues.Finally, in order to solve the problem of method based on HMM can only detect the input video which can only contain one event, HHMM is introduced to solve the conflict of the video segmentation and event detection. An HHMM model was developed to group shots and to recognize simultaneously events in a soccer video. The experiments show the proposed soccer video semantic analysis algorithms with good results.