Research on Event Detection of Soccer Video Based on Hidden Markov Model

For soccer video, the algorithm that using semantic shots as observation, the events as the states to construct the Hidden Markov Model (HMM) for event detection was proposed. The nine types of semantic shots such as front region, midfield, penalty area, the medium shot, the players close-up, outside audience, referee shot, slow motion playback shot and other types are used as observation. The four kinds of events to be detected, namely, the normal playing event, play suspension event, shooting event and foul event are defined as the states. According to the decoding principle of HMM, the states sequence with the maximum possibility for the input observation sequence was calculated and thus the event detection was completed. Compared with other event detection algorithms based on HMM assessment principle, the method adopted in this paper only has to construct one HMM and need less computation time. The experiment results show that our algorithm is effective.