Towards adaptive control in smart homes: Overall system design and initial evaluation of activity recognition

摘要

This paper proposes an approach for adaptive control over devices within a smart home, by learning user behavior and preferences over time. The proposed solution leverages three components: activity recognition for realizing the state of a user, ontologies for finding relevant devices within a smart home, and machine learning for decision making. In this paper, the focus is on the first component. Existing algorithms for activity recognition are systematically evaluated on a real-world dataset. A thorough analysis of the algorithms' accuracy is presented, with focus on the structure of the selected dataset. Finally, further study of the dataset is carried out, aiming at reasoning factors that influence the activity recognition performance.