The world population of people over the age of 65 is growing rapidly at a rate of 800,000 per month. As a consequence, the healthcare and medicare market already constitutes a major part of the European economy and it will only expand in years to come. Recent advances in sensor technology, cellular networks and information technology promise to improve the well-being of the elderly by assisting them in their daily activities and monitoring their health status, thus enabling them to lead their lives to a larger extent independently from healthcare institutions and their caretakers.
However, the comparatively slow adoption of such systems indicates that there are certain factors prohibiting their acceptance and use. For example, most available home-care systems monitor the health of individuals suffering from diagnosed chronic diseases (heart disease, lung disorders, diabetes, etc.); as such, they depend on customised, costly, and difficult to use health monitoring equipment, such as cardiographic monitors, and they often confine patients to their homes. Less attention has been given to monitoring and maintaining the personal wellness of those elderly people who have not been diagnosed with any serious or chronic disease and who, therefore, wish or should be encouraged to live a normal life, both inside and outside their homes. The quality of life of this larger population depends upon and may decline as a result of combined and (in some cases) not easily measurable physical and psychological factors. Existing work in the direction of monitoring and improving their lives, however, is limited to a) interactive facilities for consultation with doctors, which, while simplifying some regular examinations, do not take advantage of the full potential offered by the latest advances in information and communication technologies, and b) satellite location tracking systems, which, while allowing people suffering from memory and orientation problems to be easily traced, do not otherwise exploit the wealth of data gathered by the location sensors.
While sudden changes and, more generally, unexpected patterns in the daily behaviour of elderly people are difficult to detect via medical data only, it is very likely that i) such unexpected patterns can be detected by combining machine learning and input from a larger variety of sensors, and ii) that they can be used as signs that a certain disease, illness or health condition is getting worse or that is about to occur. Of course, detecting these early signs is far from trivial, if we consider the uniqueness of each individual's personality, behaviour, and the diverse effects of different diseases upon this behaviour. Furthermore, early detection should not come at the cost of confining the elderly to their homes, nor should it rely on the assumption that elderly users will learn to control complex equipment.
The aim of the ARCHANGEL project is to realise a holistic framework that will exhibit all these principles and test its performance in a controlled trial setting. In this context, the project aims to design, implement, and validate a cost-effective, secure (not compromising the monitored person's privacy), adaptable and interoperable framework for learning and monitoring the daily behaviour and personal routines of the elderly using advanced sensor networking, machine learning, and controlled interaction with caretakers. The resulting system will be based on off-the-shelf sensors and positioning-enabled cellular phones.
The ARCHANGEL architecture comprises two major components: the Monitoring Environment consisting of the software and hardware components that gather, format and encrypt the sensor data for the monitored individuals, and the Diagnostic Environment consisting of the software components that: 1) identify high-level events, 2) construct a behaviour model for each individual, 3) allow entering hand-crafted monitoring rules, 4) use both the behaviour models and the hand-crafted rules to detect conditions that require a caretaker's attention, and 5) extract generic patterns representing correlations between known diagnoses and aggregated behavioural models, which are then compared against individuals' behavioural models and diagnostic data to asses possible health risks. More specifically, the two major components have the following sub-components: