Implementing evidential activity recognition in sensorised homes

Article type: Research Article

Authors: Hong, Xin^{; *} | Nugent, Chris

Affiliations: School of Computing and Mathematics and Computer Science Research Institute, University of Ulster, Northern Ireland, UK

Correspondence: [*] Corresponding author. Tel.: +44 28 90368394; Fax: +44 28 70324916; E-mail: x.hong@ulster.ac.uk.

Abstract: Automated recognition of activities of daily living such as preparing meals and grooming may be considered as one of the most desirable computational functions within a Smart Home for the elderly. In our current work we present a process framework with the capability of realising evidential ontology networks for recognising activities of daily living in a single-person occupied inhabitancy. The performance of this framework has been evaluated using a publicly available data set consisting of 28 days worth of sensor data which was recorded from a single person living in an apartment. Within the paper we show how evidential inference networks of activities of daily living can be generated from the smart home and subsequently used to represent sensor evidence and activity performance. Based on exposure to the data set considered within the study the model achieved an overall class accuracy of 83.4% and timeslice accuracy of 95.7%. Previously reported attempts to classify this data based on a probabilistic approach achieved rates in the region of 79.4% and 94.5% respectively.

Keywords: Smart homes, ADL recognition, binary sensors, evidential ontology network of activity

DOI: 10.3233/THC-2011-0610

Journal: Technology and Health Care, vol. 19, no. 1, pp. 37-52, 2011