Improving Supervised Classification of Activities of Daily Living Using Prior Knowledge

The increase in life expectancy is producing a bottleneck at the entry in institutions. Therefore, telemedicine becomes a timely solution, which is largely explored to care after elderly people living independently at home. It requires identifying the behaviors and activities of the person at home, with non-intrusive sensors and to process data to detect the main trends in the health status. This paper presents the results of the study of prior introduction, in Support Vector Machine, to improve the automatic recognition of Activities of Daily Living. From a set of activities, performed in the experimental smart home in Grenoble, the authors obtained models for seven activities of Daily Living and tested the performances of this classification with introduction of spatial and temporal priors. Eventually, different results are discussed.