Semi-supervised adapted HMMs for unusual event detection

摘要: We address the problem of temporal unusual event detection. Unusual events are characterized by a number features (rarity, unexpectedness, and relevance) that limit application traditional supervised model-based approaches. propose semi-supervised adapted hidden Markov model (HMM) framework, in which usual models first learned from large amount (commonly available) training data, while Bayesian adaptation an unsupervised manner. The proposed framework has iterative structure, adapts new at each iteration. show such can problems due to scarcity data difficulty pre-defining events. Experiments on audio, visual, audiovisual streams illustrate its effectiveness, compared with both baseline methods.