Knowledge-driven user activity recognition for a smart house. Development and validation of a generic and low-cost, resource-efficient system

Our core interest is the development of autonomous and socially interactive robots that may support elderly users at home as part of a smart home, i.e. a home equipped with a sensor network that may detect activities of daily living such as preparing food in the kitchen, having meal in the living room, watching the television, etc. The current paper focuses on showing the design and implementation of a low-cost, resource-efficient activity recognition system that can detect user activities without the necessity of collecting a large dataset to train the system. Based on common-sense knowledge from activities of daily living, we generated a set of rules for defining user's activities in a home setting. These rules can be edited and adapted easily in order to accommodate different environments and daily life routines. The approach has been validated empirically with a pilot study in the University of Hertfordshire Robot House. The paper presents results from a study with 14 participants performing different daily life activities in the house. The results are promising, and future work will include the integration of this system in a Smart House used for Human-Robot Interaction studies. This may help develop contextaware robot companions capable of making better decisions to support users in their daily activities.