Low-Cost Non-Wearable Fall Detection System Implemented on a Single Board Computer for People in Need of Care

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Falls are a significant risk for individuals needing care, leading to severe health consequences.
• Existing fall detection systems are often costly, wearable, or raise privacy concerns.

Purpose of the Study:

• To propose an affordable, non-wearable artificial vision system for detecting falls in people needing care.
• To develop a system utilizing deep learning on a Raspberry Pi with a camera for real-time fall detection.

Main Methods:

• Implemented a Convolutional Neural Network (CNN) on a Raspberry Pi 4.
• The CNN classifies human poses into five states: fallen, crouching, sitting, standing, and lying down.
• Alerts are sent via Telegram upon fall detection.

Main Results:

• The system achieved high performance metrics: 96.4% precision, 96.6% specificity, 94.8% accuracy, and 93.1% sensitivity.
• Demonstrated effectiveness across various conditions including different outfits, lighting, and distances.
• Showcased a favorable balance between system performance and cost.

Conclusions:

• The proposed system offers an effective and affordable solution for fall detection, particularly beneficial for resource-limited regions.
• Addresses privacy concerns by not recording video and only sending images during fall events.
• Contributes to timely intervention, potentially reducing fall-related fatalities and improving care for vulnerable populations.