A Graph-Attention-Based Method for Single-Resident Daily Activity Recognition in Smart Homes

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Smart home systems are crucial for ambient-assisted living.
• Recognizing daily human activities is essential for these systems.
• Inferring activities from sensor data presents challenges due to varying time intervals.

Purpose of the Study:

• To introduce a novel deep learning framework for human activity recognition in smart homes.
• To address the challenges of inferring activities from sensor observation sequences with irregular time intervals.

Main Methods:

• Developed a novel deep learning framework: time-oriented and location-oriented graph attention (TLGAT) networks.
• Utilized embedding technology to convert sensor observations into feature vectors.
• Modeled sensor observation sequences as fully connected graphs to capture temporal and location correlations.

Main Results:

• The TLGAT networks demonstrated favorable performance in activity recognition.
• The method proved effective across diverse experimental setups and sensor event sequence lengths.
• The framework successfully facilitated feature representation through inter-sensor weighting operations.

Conclusions:

• The proposed TLGAT network is a promising approach for human activity recognition in smart home environments.
• This deep learning framework enhances ambient-assisted living by accurately inferring daily activities.
• The method's ability to handle temporal and location correlations offers robust activity recognition.