MICAR: multi-inhabitant context-aware activity recognition in home environments.

Area of Science:

• Computer Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Sensor-based Activities of Daily Living (ADL) recognition is crucial for smart-home healthcare monitoring.
• Existing methods struggle with multi-inhabitant scenarios and the challenge of data association.
• Supervised learning approaches require extensive labeled data, which is impractical for multi-resident settings.

Purpose of the Study:

• To develop a novel approach for multi-inhabitant ADL recognition in smart homes.
• To address the data association problem by linking sensor events to specific individuals.
• To overcome the limitations of supervised learning by utilizing semi-supervised techniques and reducing reliance on labeled data.

Main Methods:

• Proposed MICAR, a multi-inhabitant ADL recognition approach combining semi-supervised learning and knowledge-based reasoning.
• Implemented semantic reasoning for data association, integrating contextual information (posture, location) with sensor events.
• Utilized an incremental classifier with a cache-based active learning strategy for continuous improvement.

Main Results:

• MICAR demonstrated reliable recognition of individual and joint ADLs in a multi-subject setting (up to 4 participants).
• The approach effectively solved the data association problem by linking sensor events to the correct inhabitant.
• The active learning strategy required a significantly low number of queries to improve classifier performance.

Conclusions:

• MICAR offers a robust solution for multi-inhabitant ADL recognition in smart homes.
• The combination of semantic reasoning and semi-supervised learning effectively handles data association and labeling challenges.
• This approach advances smart-home healthcare applications by enabling accurate monitoring of multiple individuals.