[home] [Personal Program] [Help]

---

Session: Poster session II
Session starts: Thursday 24 January, 16:00

---

Marc Mertens (MOBILAB: Biosciences and Technology Department, Thomas More Kempen, Belgium)
Glen Debard (MOBILAB: Biosciences and Technology Department, Thomas More Kempen, Belgium)
Bert Bonroy (MOBILAB: Biosciences and Technology Department, Thomas More Kempen, Belgium)
Els Devriendt (Center for Health Services and Nursing Research, KU Leuven, Belgium)
Koen Milisen (Center for Health Services and Nursing Research, KU Leuven, Belgium)
Jos Tournoy (Geriatric Medicine, University Hospitals Leuven & Department of Clinical and Experimental Medicine, KU Leuven, Belgium)
Jesse Davis (Department of Computer Science, KU Leuven, Belgium)
Tom Croonenborghs (MOBILAB: Biosciences and Technology Department, Thomas More Kempen, Belgium)
Bart Vanrumste (MOBILAB: Biosciences and Technology Department, Thomas More Kempen, Belgium)

Abstract:
Due to the ageing population, more elderly people will depend on the support of others. In order to improve the quality of care and life and to keep the cost of health care sustainable, new ways to allow the elderly to live independently in their own home environment need to be considered. The aim of our research is to develop and evaluate a system that can automatically detect the activities of daily living (ADL) of older persons living alone at home. This automatic detection is based on measurements with contactless sensors that are installed in the home environment of the older person. The results can be used to assess the self reliance of the person monitored. In this research we deploy sensors that can measure the consumption of public utilities (electricity, water and gas), movement sensors and video cameras. This abstract describes the results of the first step towards recognition of ADL using electricity sensoring at one central point at the electricity cabinet. By identifying which electrical appliance is being turned on or off, we can deduce the ADL the person is performing with a certain probability. Recently there has been interest in the automatic classification of electrical appliances [1], mainly in the field of energy management. We try to improve results by using extended features describing the electrical current signal and by applying machine learning techniques. We divide the appliances into four categories, based on their electrical architecture: lighting (e.g., fluorescent, incandescent, halogen), motored appliances (e.g., hairdryers, mixers) heating (e.g., halogen heaters) and electronics appliances (e.g., TV sets, radios, computers) The features we use to discriminate between appliance classes include power, phase shift with respect to mains voltage, harmonics, correlation with a pure sine wave, etc. We measured the electrical current profile of several commonly used appliances (e.g., microwaves, hairdryers, lighting) which can be linked to an ADL (e.g., cooking, eating). Measurement specifications were 5kS/s sampling speed and 12 bit resolution. Our training set consists of 106 measurements covering 16 appliances. Using the simple J48 tree classifier in the machine learning toolkit Weka, we correctly classified 95 out of 106 instances. This test is a good first step in appliance discrimination towards detecting performed ADL, especially when it will be combined later on with information such as position and posture information. The next step in our research is working with combined signals. When several appliances are powered simultaneous, we need to first decompose the electrical current signal into the discrete current profiles.