A novel approach to creating artificial training and test data for an HMM based posture recognition system

Abstract

Demographic change in the next few years will lead to a pronounced disparity in generation distribution. Hence there is a need to develop intelligent systems to support and maintain the autonomy of the elderly at home. A high priority in this case assumes the preparation-free acquisition of vital signs and patient parameters in long-term monitoring systems to detect early changes or deterioration in health. It is thus possible to initiate treatment of a disease at an early stage. One way to carry out a long-term monitoring of vital signs at home is based on the functionalization of furniture, for example, through the use of suitable sensors in chairs [1] and beds [2, 3] to derive various patient parameters. In addition to monitoring basic parameters, e.g. the heart rate and respiratory activity, it is also possible to access information regarding motion or sleep patterns by means of pattern recognition systems. In addition to the challenge of building a suitable pattern recognition system there is a need for corresponding training data to create reference patterns. Typically, the necessary sensor data for the reference pattern training is generated in time-consuming sessions with real people. In this paper, a novel approach is presented, which provides a multi-stage model to create artificial training or test data. The model can be used as a supporting tool in the development of posture recognition systems and to create artificial data for training and testing.