Ubiquitous sensor-based human behaviour recognition using the spatio-temporal representation of user states

Abstract

There has been much research on location-based context-aware applications. However, any description of a person's activities must include a temporal aspect as well as a location aspect. Therefore, it is important when creating enhanced user activity support systems to consider the user's context in terms of spatio-temporal constraints. In this paper, we propose a user activity support system that employs a state sequence description scheme to describe the user's context. In this scheme, each state is described as a spatio-temporal relationship between the user and objects. Typical sequences of states are stored as models of activities performed by a user. Each segment of user activities measured by the sensors and the Radio Frequency Identification tags (RFID tags) is classified into a state by using a decision tree constructed by the machine learning algorithm called C4.5. The user's context is then obtained by matching the detected state series to a stored task model. To validate this system, we have developed an experimental house containing various embedded sensors and RFID-tagged objects. Having evaluated the performance of the proposed system, we conclude that our system is an effective way of acquiring the user's spatio-temporal context.