4th International ICST Conference on Pervasive Computing Technologies for Healthcare

Research Article

Using duration to learn activities of daily living in a smart home environment

Download738 downloads
        author={Shuai Zhang and Sally McClean and Bryan Scotney and Priyanka Chaurasia and Chris Nugent},
        title={Using duration to learn activities of daily living in a smart home environment},
        proceedings={4th International ICST Conference on Pervasive Computing Technologies for Healthcare},
        keywords={probabilistic learning; duration; reasoning; ADL;smart home},
  • Shuai Zhang
    Sally McClean
    Bryan Scotney
    Priyanka Chaurasia
    Chris Nugent
    Year: 2010
    Using duration to learn activities of daily living in a smart home environment
    DOI: 10.4108/ICST.PERVASIVEHEALTH2010.8804
Shuai Zhang1,*, Sally McClean1,*, Bryan Scotney1,*, Priyanka Chaurasia1,*, Chris Nugent2,*
  • 1: School of Computing and Information Engineering, University of Ulster, Coleraine, Northern Ireland
  • 2: School of Computing and Mathematics, University of Ulster, Newtownabbey, Northern Ireland
*Contact email: s.zhang@ulster.ac.uk, si.mcclean@ulster.ac.uk, bw.scotney@ulster.ac.uk, chaurasia-p@email.ulster.ac.uk, cd.nugent@ulster.ac.uk


Recognition of inhabitants' activities of daily living (ADLs) is an important task in smart homes to support assisted living for elderly people aging in place. However, uncertain information brings challenge to activity recognition which can be categorised into environmental uncertainties from sensor readings and user uncertainties of variations in the ways to carry out activities in different contexts, or by different users within the same environment. To address the challenges of these two types of uncertainty, in this paper, we introduce the innovative idea of incorporating activity duration into the framework of learning inhabitants' behaviour patterns on carrying out ADLs in smart home environment. A probabilistic learning algorithm is proposed with duration information in the context of multi-inhabitants in a single home environment. The prediction is for both inhabitant and ADL using the learned model representing what activity is carried out and who performed it. Experiments are designed for the evaluation of duration information in identifying activities and inhabitants. Real data have been collected in a smart kitchen laboratory, and realistic synthetic data are generated for evaluation. Evaluations show encouraging results for higher-level activity identification and improvement on inhabitant and activity prediction in the challenging situation of incomplete observation due to unreliable sensors compared to models that are derived with no duration information. The approach also provides a potential opportunity to identify inhabitants' concept drift in long-term monitoring and respond to a deteriorating situation at as early stage as possible.