Opportunistic Activity Recognition Methodologies

The technological environment that surrounds us nowadays - evolved by the invisible integration of digital artifacts into everyday objects - is equipped with mobile and interconnected technology delivering sensor data characterizing our current activities and more generally the context. For example, today's smart phones can be regarded as multi-sensor platforms delivering high quality multi-modal environmental data. This aspect reverses the traditional approach of human activity recognition, since the artificial and obtrusive deployment of sensors in the environment, on objects, or on people's bodies can be regarded as obsolete. This thesis presents a novel approach for recognizing the activities (and more generally the context) of individuals, by taking advantage of currently available sensor devices in an opportunistic way, without the effort of initial sensor definition and deployment. Thus, the sensing infrastructure in an opportunistic system cannot be defined and fixed before operation, which brings about a novel and innovative scientific area with challenging aspects that compose the research conducted in this thesis: (i) utilization of spontaneously available heterogeneous sensors without prior knowledge of the physical, working, and deployment characteristics and (ii) the handling of a spontaneous and dynamic sensing infrastructure. (...)