Unsupervised daily routine and activity discovery in smart homes

Area of Science:

• Smart Home Technology
• Artificial Intelligence
• Gerontology

Background:

• Smart homes are crucial for remote health monitoring through activity recognition.
• Current supervised methods for activity recognition are costly and time-consuming due to manual labeling.

Purpose of the Study:

• To present an unsupervised method for discovering daily routines and activities in smart home residents.
• To enable automatic identification of patterns indicative of functional decline in the elderly.

Main Methods:

• Utilizes a Markov chain to model resident locomotion patterns throughout the day.
• Discovers macro-level daily routine clusters.
• Identifies micro-level room-specific activities within each routine cluster.

Main Results:

• Successfully clusters daily routines at a macro level.
• Identifies specific room-level activities within identified routines.
• Provides a foundation for automated health monitoring and intervention.

Conclusions:

• The unsupervised approach offers a practical alternative to supervised methods for smart home activity recognition.
• Automatic discovery of routines and activities can support early detection of elderly functional decline.
• Timely interventions can be suggested based on identified activity patterns.