Edge Computing Transformers for Fall Detection in Older Adults

Area of Science:

• Gerontology
• Biomedical Engineering
• Wearable Technology

Background:

• Increasing life expectancy presents challenges, notably a higher risk of falls in older adults.
• Existing fall detection systems suffer from server delays, high false positives, poor wearability, and high costs.
• These limitations impact individual health, quality of life, and healthcare systems.

Purpose of the Study:

• To develop a reliable, wearable, and cost-effective fall detection system for older adults.
• To address the limitations of current fall detection technologies.
• To enable real-time fall detection through an embedded system.

Main Methods:

• A wearable device incorporating an Inertial Measurement Unit (IMU) was developed.
• A hybrid algorithm combining a Threshold-Based Algorithm (TBA) and a low-parameter Transformer neural network was implemented.
• The system enables real-time fall detection without continuous server communication.

Main Results:

• The system achieved high performance metrics: 95.29% accuracy, 93.68% specificity, and 96.66% sensitivity.
• It demonstrated significant efficiency by using only 0.38% of the trainable parameters compared to other approaches.
• Real-time detection capabilities were successfully implemented.

Conclusions:

• The proposed system offers a reliable, wearable, and cost-effective solution for fall detection in older adults.
• The hybrid algorithm and embedded design overcome limitations of existing systems.
• This technology has the potential to enhance the safety and quality of life for the aging population.