关键词: 时空上下文, 多媒体本体, 语义分级, 语义标注, 异常检测, 多层隐马尔科夫模型, 悲观情感模型, ,

Abstract: The important and difficult problem in smart home monitoring system is to detect abnormal activities of empty nest elder or devices and give an alarm in time.Multi-Model Joint Sensors（MJS） technique is used to capture the spatio-temporal context，which is processed by all kinds of perceptual components to obtain the action of occupants.Improved Hierarchical Hidden Markov Model（HHMM） is used to abstract discrete action into human’s high-level behavior—event.The representation models for occupant normality are constructed from large number of spatio-temporal data and used as classifiers that recognize normality or abnormality，to detect occupant’s abnormal behavior.In order to describe context information，a set of new Multi-Media Ontology （MMO） is designed by utilizing the method of semantic-layering and semantic-abstracting step by step，and used to annotate and reason about media information in smart monitoring system.A Pessimistic Emotion Mode（PEM） is improved，which is used to analyze multi-interleaving event of multi-active devices in home，to solve the problem that video component has difficulty in detecting state change of active devices.The experiences demonstrate that the solution has good performance in detecting abnormality and sending instant alarms.

Key words: spatio-temporal context, Multi-Media Ontology（MMO）, semantic-layering, semantic annotation, abnormality detection, Hierarchical Hidden Markov Model（HHMM）, Pessimistic Emotion Model（PEM）