Towards real world activity recognition from wearable devices

Supporting people in everyday life, be it lifestyle improvement or health care, requires the recognition of their activities. For that purpose, researches typically focus on wearable devices to recognize physical human activities like walking whereas smart environments are commonly the base for the recognition of activities of daily living. However, in many interesting scenarios the recognition of physical activities is often insufficient whereas most smart environment works are restricted to a specific area or one single person. Moreover, the recognition of outdoor activities of daily living gets significantly less attention. In our work, we focus on a real world activity recognition scenario, thus, practical application including environmental impact. In this context, we rely on wearable devices to recognize the physical activities but want to deduce the actual task, i.e., activity of daily living by relying on background and context related information using Markov logic as a probabilistic model. This should enable that the recognition is not restricted to a specific area and that even a smart environment could be more flexible concerning the number of sensors and people. Consequently, a more complete recognition of the daily routine is possible which in turn allows to perform behavior analyses.