Fall Monitoring With Single IMU: A Large-Scale Dataset and a Novel Dual-Branch Network.

Area of Science:

• Biomedical Engineering
• Gerontology
• Wearable Technology

Background:

• The growing elderly population faces significant health risks from fall accidents.
• Proactive fall prevention through pre-impact fall detection (PIFD) is crucial in healthcare monitoring.
• Wearable sensors, particularly Inertial Measurement Units (IMUs), offer a promising avenue for PIFD.

Purpose of the Study:

• To construct and release a large-scale dataset (FallTL) for fall and activities of daily living (ADL) detection.
• To develop and evaluate a novel Spatial-Temporal Attention Network (STA-Net) for PIFD using IMU data.
• To investigate the feasibility of accurate PIFD with minimally obtrusive, single-location sensor setups.

Main Methods:

• Construction of the FallTL dataset, encompassing falls and ADLs from multiple body segments using IMUs.
• Development of STA-Net, a dual-branch neural network with temporal and spatial attention mechanisms for IMU-based PIFD.
• Cross-dataset and cross-subject validation of STA-Net against state-of-the-art methods.

Main Results:

• STA-Net achieved superior performance and comparable lead times in PIFD across multiple datasets under cross-subject validation.
• The model outperformed existing state-of-the-art baselines in fall detection accuracy and timeliness.
• Analysis confirmed that effective PIFD is achievable with sensors placed on a single body segment.

Conclusions:

• The FallTL dataset and STA-Net provide valuable resources for advancing research in wearable-based fall detection.
• Accurate and robust pre-impact fall detection is feasible using single-location, minimally obtrusive wearable sensors.
• These findings have significant practical implications for developing effective wearable fall monitoring systems for the elderly.