Fusion of multiple sensors sources in a Smart Home to detect scenarios of activities in ambient assisted living

This work takes place within the framework of Smart Homes, with the goal to monitor the activities of elderly people, living independently at home, in order to continuously assess their level of activity and therefore their autonomy. A method is proposed for the selection of a range of sensors and for multiple data fusion. The system was evaluated on 7 young and 4 elderly healthy subjects who performed scenarios of daily activities (sleeping, eating, walking, and transfer) within a controlled environment. These activities were correctly classified with an overall sensitivity and specificity of 67.0% (out of 267 activities) and 52.6% (502) for the group of young people, and of 86.9% (222) and 59.3% (492) for the elderly group. The results were better with activities commonly performed in a dedicated location (i.e., taking meals in the kitchen, toileting in the bathroom). The results are acceptable with a reduced set of sensors although numerous and/or more informative sensors (i.e., video, sound detection, sensitive floors, etc.) give higher results at the cost of more cumbersome and costly systems, difficult to deploy in a private home and eventually more intrusive.