Enhancing elderly care: Efficient and reliable real-time fall detection algorithm.

Area of Science:

• Computer Vision
• Biomedical Engineering
• Artificial Intelligence

Background:

• Falls are a major public health risk, particularly for the elderly, necessitating effective detection systems.
• Vision-based fall detection offers a non-invasive and affordable alternative to wearable sensors.
• Current algorithms face challenges with computational demands, limiting on-device speed and performance in complex environments.

Purpose of the Study:

• To develop a real-time fall detection algorithm with low computational requirements.
• To address the limitations of existing systems in terms of speed and performance in complex scenes.
• To create an efficient and accurate fall detection solution using readily available hardware.

Main Methods:

• Implemented a real-time fall detection algorithm utilizing a single webcam.
• Combined background subtraction with the BlazePose human pose estimation model for optimized precision and efficiency.
• Employed biomechanical features derived from BlazePose key points within a random forest model for fall event classification.

Main Results:

• Achieved 89.99% accuracy and 29.7 frames per second (FPS) on standard fall detection datasets.
• Demonstrated strong generalization and robustness across diverse scenarios.
• The algorithm operates efficiently on a standard laptop CPU, indicating low computational demands.

Conclusions:

• The developed algorithm offers a practical solution for real-time fall detection.
• Its low computational power makes it suitable for integration into small-scale medical monitoring equipment.
• The system holds significant potential for enhancing digital health applications and patient safety.