Automatic emergency detection using commercial accelerometers and knowledge-based methods

This paper focuses on the challenge of automatical detecting an emergency, e.g., a fall by an elderly person, and to generate an alert such as a phone call or sending a SMS to a relative as fast as possible. The presented system only needs one single triaxial accelerometer. The algorithmic part uses the paradigm of knowledge-based methods. Unlike pattern recognition algorithm, knowledge-based methods strictly separate between the so-called knowledge base declaratively describing the knowledge about the specific domain and the so-called inference component or inference engine that tries to derive answers from the underlying knowledge base. That is to say the knowledge base can be replaced without changing the concrete inference machine. The main part of the developed algorithm to detect falls is based on a fuzzy logic inference system and a neural network. In addition, the current velocity and relative position of the person wearing the sensor are determined from acceleration data. These information can be used as further features to improve both sensitivity and specificity. The described methods were integrated into the telemedical system described.