Partitioning time series sensor data for activity recognition

Monitoring activities of daily living is one of the key functionalities expected from a Smart Living environment providing independent living services for elderly people. Simple state-change sensors have been considered to be a promising sensing technique to observe the environment and consequently provide the data required to form the basis to infer high-level behaviours. From a data analysis point of view, the challenge is how to recognise and detect activity behaviours from low level sensor data over time. In this paper we present a novel approach to partition sensor data and identify the activity undertaken within each sensor data segment. The approach developed was tested on a dataset collected from a single person living in an apartment during a period of 28 days. The results show that our approach can not only accurately recognise annotated activities but also has the ability to identify non-recorded activities.