Research On Detecting Abnormality For Empty Nest Elder In Smart Monitoring System

With the development of economy and society, empty nest family has become main habitation fashion of the elder in China. Detecting abnormal activities of empty nest elder or devices and alarming in time is an important and difficult problem in smart home monitoring system. The key lies in its perception on the environment and understanding the context.The thesis analyses and summarizes the characteristics and shortcomings of previous intelligent monitoring system, researches how to get and organize the environment context (knowledge) needed for the abnormality detection on ubiquitous conditions, and proposes a new prototype of smart monitoring system. It can detect abnormality of the elderly in empty nest, and take corresponding measures according to the type of abnormality and the level of reliability.First, in order to represent 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 for bridging "semantic gap" between perceivable media world and high-level conceptual world, it provide semantic knowledge for events reasoning.Second, in order to implement unsupervised learning of human behavior, in this paper, a modified hierarchical hidden markov model is used to construct a model of resident's normal behavior, which is used as the classifier of normal and abnormal behavior, consequently achieve the unsupervised detection of abnormal behavior of residents. Simulation results show that, it has better ability to identify than the HMM model commonly used in previous studies when it used in the same sample.Third, for the hard-to-detect problem of abnormal event under video surveillance equipment, a Pessimistic Emotion probability Mode (PEM) is improved to analyze multi-interleaving event of multi-active devices in home. The reliability and rationality is obviously improved for abnormality detection of multi-devices.Finally, the paper sums up our work, and discusses how to improve the system in the future.