Event-driven Activity Recognition from Heterogeneous Sensor Data

Abstract

A smart environment should monitor and recognize the activities of people living within the space in order to provide timely support for safety, comfort, and convenience. This research proposes an approach to activity recognition based on data collected from sensors embedded in the environment. A variety of simple sensors are used to detect different aspects of the human behaviors. This paper presents event-driven activity recognition using an event-action-activity model. A multi-agent architecture is proposed to enable context-aware services based on user activities gleaned from aggregating and interpreting sensor data. We conducted experiments to evaluate the systemís performance in recognizing a number of routine everyday activities. Three applications in personal assistance are described in this paper.