This paper details an end-to-end system for recognising activity in smart homes for e-care. It discusses the hardware options to be considered when designing the smart home, and the particular decisions takes at the eWALL system. It then considers the necessary signal processing algorithms that turn measurements into metadata describing the context of the care recipient. Since activity recognition implementation and testing need long-term measurements and metadata, a realistic simulator is also built for providing input to the activity recognition module. The activity recognition algorithm utilises two models, one for location estimation and another for activity estimation in the given location. They both give correct recognition for 96 and 94% of the time.
Keywords: Activity recognition, Smart home, Home sensing, Context extraction