Towards the combination of statistical and symbolic techniques for activity recognition

Techniques for activity recognition are fundamental components of any context-aware system. Indeed, precise knowledge of the user´s current activity is necessary in order to thoroughly tailor services to the user´s context. To this aim, in the last years many research efforts have been devoted to devise statistical techniques for recognizing basic physical activities based on data retrieved from body- worn sensors. An intriguing research issue is how to integrate those statistical techniques with symbolic ones in order to i) refine statistical predictions, and ii) derive more complex activities. The intuition behind this research direction is that the current user´s context (e.g., artifacts and persons in the user´s surrounding environment) may give useful hints about the possible activities performed by the user herself. Given an ontology that models the addressed scenario, those hints can be automatically derived through ontological reasoning and used to refine the prediction of the statistical classifier. Moreover, ontological reasoning can be performed to derive complex activities described in terms of simpler activities and symbolic constraints about the user´s context.