Distributed agent system for behavior pattern recognition

This paper describes a system that is being developed for providing services to support elderly residents in assisted living facilities. Passive RFID tags and readers are used as the basic sensor modality to distinguish people and objects. An intelligent multiagent system with a distributed topology enables local decision making without the need for a centralized server or database. Rather than use predefined decision rules, we employ agents that extract behavior pattern of the residents from sensor data, use these patterns to identify abnormal events, and trigger actions to assist the residents. Because the system is implemented on a resource constrained platform, a light weight clustering algorithm is utilized to generate data models of the residents´ daily activities. Finally, a set of local Markov chains enables future locations of the residents to be predicted.