Using Graphs to Perform Effective Sensor-Based Human Activity Recognition in Smart Homes.

Area of Science:

• Computer Science
• Artificial Intelligence
• Ubiquitous Computing

Background:

• Human Activity Recognition (HAR) is crucial for smart homes, ambient intelligence, and assisted living.
• Existing HAR systems face challenges like data variability, sparsity, and noise.
• Current state-of-the-art HAR methods require pre-segmentation of sensor data, limiting real-world deployment.

Purpose of the Study:

• To propose a novel graph-guided neural network approach for HAR in smart homes.
• To overcome the limitation of pre-segmentation in existing HAR systems.
• To enable automated analysis of continuous sensor data streams without manual intervention.

Main Methods:

• Developed a graph-guided neural network that learns explicit co-firing relationships between sensors.
• Utilized a data-driven approach to learn an expressive graph structure representing the smart home sensor network.
• Employed attention mechanisms and hierarchical pooling of node embeddings to map discrete sensor measurements to a feature space.

Main Results:

• The proposed graph-guided neural network significantly outperforms state-of-the-art HAR methods on CASAS datasets.
• Demonstrated superior performance across multiple datasets, indicating robustness and effectiveness.
• Achieved large margins of improvement compared to existing approaches.

Conclusions:

• The novel graph-guided neural network effectively performs HAR without requiring pre-segmentation of sensor data.
• This approach enhances the practical applicability of HAR systems in real-world smart home environments.
• The findings represent a significant step towards more autonomous and reliable smart home technologies.