Elderly People Living Alone: Detecting Home Visits with Ambient and Wearable Sensing.

Ubiquitous computing techniques are enabling the possibility to provide remote health care services to elderly citizens. In such systems, daily activities are extracted from raw sensor signals, based on which users? health status can be inferred. Due to the ambiguity of raw sensor signals, it is challenging to distinguish the number of people in the ambient, and most such systems assume user live alone. We present an algorithm to automatically detect home visits to elderly people living alone, using an ambient and wearable sensing network. We use visiting reports from caregivers as partially labeled positive data, and conduct statistical analysis to gain insights of visit events in terms of raw sensor data, based on which a set of features are extracted. A one-class support vector machine is trained on a small set of positive data from one user, and tested on five installations. Experimental results show that our algorithm can correctly detect 58%-83% of the labeled visits using only the ambient sensors. The detection rate is improved by incorporating the activity data from Fitbit activity tracker, i.e., with which 75%-87% visiting events are detected. Our system is implemented and tested in the context of a real life health care system.