A Situation-aware Wearable Computing System for Human Activity Recognition

Wearable devices support the detection of human activities without the expensive and obstructive presence of environmental sensors like cameras, and without the need to constrain the users to remain in controlled environments. Human activity recognition supports many applications in several domains like safety, security, surveillance, health monitoring. To concretely and effectively support such applications, however, the simple recognition of the activities alone is not enough. What is really needed is a deep understanding of the situations in which the activities are performed to estimate the potential risks and opportunities for the application goals. In such a context, built atop a reference architecture for situation-aware wearable computing systems, this work proposes a situation-aware human activity recognition system based on hidden Markov models and the Endsley's model of situation awareness. The proposed system, implemented as a set of microservices, is flexible and adaptive as it is able to exploit the number of sensors that are available, without the need to reconfigure the system itself, thus providing the users with a best-effort approach to estimate the activities and the situations using all the currently available data.