Article Preview
TopIntroduction
The ability of an elderly person to live independently in their own home depends on their autonomy to perform the basic actions involved in daily living: to transfer to/from bed and in/out of a chair, to move around and out of the flat, to wash, to use the toilet, etc. Actually, there is a direct relationship between the number of activities performed daily by the person and their level of autonomy.
The level of autonomy of a fragile person is currently estimated by geriatricians with using manual scales. One such scale, the Activities of Daily Living (ADL) (Katz, 1963), involves 6 tasks (bathing, dressing, toileting, transferring, continence and eating) which are individually assessed by the professional as being “autonomous,” “partially autonomous,” or “non autonomous” for a given patient. This assessment is operator-dependent and cannot be performed with sufficient frequency to detect the slow trends characteristic of a loss of autonomy. Thus, there is a need for a system which can perform this evaluation in a more objective manner and on a more regular basis.
The Eureka project “DynaPort” (Van Lummel, 1996) proposed a method to monitor the activities of daily living using a wearable accelerometer sensor; it did not aim at detecting the criteria included in the ADL scale. Glascock (2000) used several sensors fixed on household appliances and furniture (e.g., fridge or cupboard doors) to detect some tasks on the scale Instrumented Activities of Daily Living (IADL) (food preparation, housekeeping, use of the telephone, etc). Duchene (2004) proposed a data fusion method to extract patterns which present similarities, in multidimensional and heterogeneous signals. She considered 4 parameters (displacement, postures, activity level and heart rate) and extracted similar patterns using different metrics. However, the current activity was not identified precisely.
At present, no system proposes to automatically and continuously detect ADLs. We therefore seek to address this goal, while using a reduced set of sensors in order to keep the solution practicable and cost effective. In a previous study we developed a kinematic sensor, called “Actimometer” (Noury, 2002; Barralon, 2005), fixed onto the chest of the person to detect the kinematics (postures, transfers, walking) of the subject, and we placed presence detectors in the rooms of our experimental flat to determine the patient’s spatial context (kitchen, bedroom, etc.).
In the first part of this paper we present the framework of multi-sensor fusion, then the selection of various sources of information and eventually the method we proposed for the data fusion. In the second part we describe our experimental protocol and results obtained with two groups of people, young and elderly.