Abstract
This paper describes a preliminary study of an activity monitoring system for a residential house based on Bayesian networks. For the demonstrative purpose, a proof-of-concept system using vibration measurement has been proposed, in which specific physical features of the vibration that may indicate either daily activities of residents or other disturbances such as traffic-induced vibration are extracted from the collected data. Vibration data is classified into two groups based on these features, and the attribution degree was defined based on one-class support vector machines associated with the extracted data classes. A Bayesian Network is designed as the data model that represents the relevance of the data classes to external variables such as time and in/out status. Finally an abnormality index has been proposed and investigated based on the likelihood evaluated on the trained Bayesian network.
