Detection of Abnormal Action Using Image Sequence for Monitoring System of Aged People

Abstract

Japan is now a serious aging society and this problem will be more serious in the near future. Therefore, engineering support for aged people is required. As an important example of such support, a monitoring system for aged people is proposed in this paper. This system automatically learns aged people's usual actions in a few days and detects his/her abnormal action by observing the room with a simple camera such as CCD camera.
At the learning stage, Self-Organizing Map (SOM) is used to realize automatic learning. For the detection method, Eigenspace Method is used which is excellent in compression of image data and calculation of the correlation among images. Parametric Eigenspace Method(PEM) is also used to detect abmormalities such as the speed of the action. Some results show the effectiveness of the proposed method.