In this paper, the authors propose software architecture to monitor elderly or dependent people in their own house. Many studies have been done on hardware aspects resulting in operational products, but there is a lack of adaptive algorithms to handle all the data generated by these products due to data being distributed and heterogeneous in a large scale environment. The authors propose a multi-agent classification method to collect and to aggregate data about activity, movements, and physiological information of the monitored people. Data generated at this local level are communicated and adjusted between agents to obtain a set of patterns. This data is dynamic; the system has to store the built patterns and has to create new patterns when new data is available. Therefore, the system is adaptive and can be spread on a large scale. Generated data is used at a local level, for example to raise an alert, but also to evaluate global risks. This paper presents specification choices and the massively multi-agent architecture that was developed; an example with a sample of ten dependant people gives an illustration.
TopIntroduction
In the year 2020, more than 20% of the industrial countries population will be over 65 (Mc Morrow, 2004) (Figure 1). This leads to organizational and financial problems for the healthcare systems. To reduce costs, and to improve their welfare, home-monitoring systems can bring solutions to help elderly people staying at home.
Figure 1. Population ages 65 and over (% of the total population)
Our project takes place in this context. It aims to help healthcare professionals by increasing the number of elderly people looked after in their home with an adaptive and non-intrusive remote assistance. Our approach tackles the home monitoring issue in a collective and cooperative way. It is based on a study of several state-of-the-art home care systems, and differs from them in that it is centred on groups instead on individuals.
We collect individual monitoring with the aim of detecting global behaviour patterns. In this article, we focus on how communities emerge from a multi-agent classification process. In such context, a multi-agent architecture is suitable because of the distribution of input data and the need for scalability: adding or removing sensors, without impacting the system’s functioning.
Patterns collected at a micro level are used to estimate the state of elderly people, to link them to their community at a macro level, to try to forecast the evolution of their activity and also to evaluate global risks (Figure 2).
Figure 2. Risk detection and alert generation
This paper is divided into three sections. First, we present many home-monitoring systems and show how our system is different. Secondly, the multi-agent classification method used in our system is described. We tested this method with a small group of elderly. This experimentation is described in the last section. We also give the first results and we highlight the future developments.