Activity recognition enhancement based on ground-truth: Introducing a new method including accuracy and granularity metrics

Abstract

The uncertainty associated with existing sensing technologies and reasoning methods affects the outcome of the activity recognition process (e.g., accuracy, precision, granularity). The activity recognition process is even challenging when switching from laboratory towards real deployments, where scenarios are not predefined and more complex. Therefore we propose a novel method to improve the activity recognition outcome, by finding a proper balance between accuracy and granularity. The method has been validated through the deployment of UbiSMART (an AAL framework) in 45 scenarios of ageing in place. We discuss in this paper our method and the validation results.