Human activity recognition using wireless ambient sensor networks

Human activity recognition in home settings provides great facilities in ambient assisted living applications. With continuous and long term monitoring, daily routines of the residents can be inferred and any abnormal situation which can be an indicator of a disease can be detected. Furthermore, health professionals can be informed in advance in such situations. Recent advances in sensor network technologies enable researchers to utilize wireless sensor networks in human activity monitoring applications. We present an ambient assisted living system which monitors the daily living of residents. For this purpose, we deployed a wireless sensor network which consists of many ambient sensors in a real house in which two residents live. Data about daily living activities of residents were collected for 30 full days accounting the privacy issues under real world conditions. Using several machine learning methods, we classified the collected data in order to model behaviours of residents and make inferences about their habits. In this thesis, we elaborate the system architecture of the wireless sensor network, share the experiences obtained during the data collection process, and the results of the classification.