An eight-camera fall detection system using human fall pattern recognition via machine learning by a low-cost android...

Area of Science:

• Gerontology
• Computer Science
• Biomedical Engineering

Background:

• Falls are a primary cause of unintentional injuries, leading to significant disability and mortality.
• Existing fall detection systems suffer from issues like battery dependence, user discomfort, high costs, and limited range.
• Non-wearable systems typically fail to detect falls beyond ten meters.

Purpose of the Study:

• To design and develop a comprehensive, house-wide fall detection system.
• To overcome the limitations of current fall detection technologies, including range and occlusion.
• To create a reliable and cost-effective solution for monitoring falls in various environments.

Main Methods:

• Utilizing a local, low-cost single-board computer for fall pattern analysis.
• Employing machine learning algorithms and crafted rules to differentiate falls from daily activities.
• Implementing multi-camera setups or high-altitude single cameras to avoid occlusion.
• Leveraging conventionally available surveillance systems for monitoring.

Main Results:

• The system successfully detects various types of falls (stumbling, slipping, fainting) at distances of 60 meters and beyond.
• Detection is effective even through transparent materials like glass and screens, and in adverse conditions like rain.
• The system accurately distinguishes true falls from non-fall events, minimizing false positives.
• Occlusion issues are mitigated through strategic camera placement.

Conclusions:

• The developed house-wide fall detection system offers a robust and extended-range solution.
• Its flexibility and low cost make it suitable for deployment in senior homes, rehab centers, and nursing facilities.
• The system can be configured for high-precision or high-recall applications, enhancing safety in high-risk areas.