Detecting deviations from activities of daily living routines using kinect depth maps and power consumption data
Journal of Ambient Intelligence and Humanized Computing
School of Science
There has been an increase in the world’s population of elderly persons who wish to live independently for as long as possible. This paper presents an unsupervised approach to help caregivers detect deviations from daily routines of elderly people living alone via inexpensive minimally intrusive sensors. The approach employs a power sensor to measure household composite power consumption and a small number of Kinect sensors in functional areas of the house to capture depth maps from the occupant’s activities of daily living (ADLs). The ADLs in an unlabelled training dataset are identified based on associating the occupant’s locations with specific power signatures on the power line. This training dataset is processed in order to model key features of ADLs, including the regularity and frequency of important activities. The approach uses a novel data-driven technique to define fuzzy sets over ADL attributes tailored to the occupant’s behaviour patterns. The membership functions of these fuzzy sets are learned based on the data distribution of attributes. A set of fuzzy rules is generated to indicate the occupant’s deviation from the normal routine of ADLs in subsequent data. The outputs of this monitoring system are reports on upward and downward deviations from normal behaviour patterns in the form of both numerical and linguistic information. The assessment of these scores over a long-term can help caregivers detect the warning signs of persistent drifts from the daily routine. As a proof of concept, the proposed monitoring approach was evaluated using two datasets collected from real-life settings. The fuzzy rule set obtained from the output of the proposed membership function generation technique was able to effectively monitor the ADLs of elderly people because it could accurately distinguish periods of deviations from the routine performance of ADLs. Compared to existing monitoring techniques, the proposed method required no prior information about the appliances in use and its output was considered to be more helpful for caregivers. The fuzzy inference system in this approach was found to be robust in regard to errors when identifying ADLs as it could effectively classify normal and abnormal behaviour patterns of the occupant despite errors in the list of the used appliances. © 2019, Springer-Verlag GmbH Germany, part of Springer Nature.
Securing Digital Futures
Artificial intelligence and autonomous systems