Two-threshold energy based fall detection using a triaxial accelerometer.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Gerontology

Background:

• Elderly fall detection using accelerometers is crucial for safety.
• Current methods often struggle to balance accuracy and energy efficiency.
• Reducing fall detection failure rates remains a significant challenge.

Purpose of the Study:

• To propose a novel double threshold methodology for accurate elderly fall detection.
• To introduce two new detection features to improve performance.
• To address the limitations of existing fall detection strategies.

Main Methods:

• Developed a double threshold based methodology.
• Introduced novel features: product of sum vector magnitude and signal magnitude area, and normalized signal magnitude area over 1s windows.
• Validated the methodology on the public Mobifall dataset and a self-developed dataset.

Main Results:

• Achieved 99% accuracy on the Mobifall dataset.
• Achieved 97% accuracy on the self-developed dataset.
• Methodology demonstrated effectiveness in reducing detection failures through activity-specific analysis.

Conclusions:

• The proposed double threshold methodology offers a promising solution for elderly fall detection.
• The novel features contribute to improved accuracy and reliability.
• Activity-by-activity analysis is an effective strategy for reducing fall detection failures.