Skeleton-Based Fall Detection with Multiple Inertial Sensors Using Spatial-Temporal Graph Convolutional Networks.

Area of Science:

• Wearable technology
• Biomedical engineering
• Machine learning for healthcare

Background:

• Wearable devices are crucial for fall detection research.
• Existing systems suffer from high false positive rates.
• Human joint dependencies are underutilized in current fall recognition schemes.

Purpose of the Study:

• To propose an accurate and reliable fall detection method.
• To leverage human joint dependencies for improved recognition.
• To reduce false positives in fall detection systems.

Main Methods:

• Utilized human motion data from inertial measurement units (IMUs).
• Developed a human skeleton model based on body joint connections.
• Applied spatial-temporal graph convolutional networks (ST-GCN) for feature extraction.
• Compared ST-GCN with MLP, CNN, RNN, LSTM, TCN, TST, and MiniRocket.

Main Results:

• The ST-GCN model demonstrated superior performance over seven other algorithms.
• Achieved higher accuracy, precision, recall, and F1-score in fall detection.
• Effectiveness of hyperparameters and window size on performance was analyzed.

Conclusions:

• The proposed ST-GCN method offers a significant advancement in IMU-based fall detection.
• This approach enhances the accuracy and robustness of fall detection systems.
• Provides a valuable reference for future research in wearable fall detection.