Affiliations: Fraunhofer Institute for Experimental Software Engineering (IESE), Fraunhofer-Platz 1, 67663 Kaiserslautern, Germany | bInstitute of Medical Biometry and Informatics, University of Heidelberg, INF 305, 69120 Heidelberg, Germany
Abstract: Societal changes lead to an increase in the number of critical emergency situations in single households, resulting in the need for new concepts, such as automatic detection of physical weakness. This paper describes an approach that identifies deviations from a person's ‘normal’ behavior, specifically the absence of typical activities, based on sensor information. False negatives are avoided by using information about the behavior on different semantic levels. The approach was evaluated in a variety of controlled lab experiments. Normal behavior was learned with 120 typical activity scenarios in bathroom and kitchen. Three configurations were defined for each location and semantic level. 100 scenarios with and without emergency situations were performed. The system's responses were analyzed regarding correctness and reaction time. Hypothesis 1: The approach detects at least 80% of critical motionlessness situations correctly (sensitivity > 0.8), which is confirmed for every configuration. Hypothesis 2: The number of false alarms (false positives) is lower than 10% with the best configuration (false positive rate < 0.1), which is confirmed for 8 out of 12 configurations. The evaluation results suggest that it is possible to detect situations of physical weakness with sufficient reliability. The approach was also tested against a standard procedure with static thresholds.