Fall prediction algorithm with built-in instability metrics.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Fall prediction is challenging due to the nonlinear, time-dependent nature of falls, especially in uncontrolled environments.
• Existing camera-based fall prediction systems face limitations including privacy concerns, high costs, and the need for extensive modifications or wearable sensors.
• Current research prioritizes fall detection over fall prediction, leaving a gap in proactive fall prevention strategies.

Purpose of the Study:

• To introduce a novel artificial intelligence (AI) platform for human body posture monitoring and fall prediction.
• To develop a fall prediction system that overcomes the limitations of existing camera-based approaches, focusing on accuracy, cost-effectiveness, and privacy.
• To extract new, camera-independent features for improved fall prediction accuracy and early warning capabilities.

Main Methods:

• Utilized a 4K camera to record various fall scenarios.
• Extracted novel features including key body landmarks, centroid locations, and angular positions of body segments.
• Developed an AI platform to analyze these features for fall prediction.

Main Results:

• The AI platform achieved an approximate accuracy of 91% in predicting falls.
• Feature importance analysis confirmed the significance of the extracted features in enhancing prediction.
• The system can predict falls up to two seconds before they occur, offering a significant improvement over existing single-camera systems.

Conclusions:

• The proposed AI platform offers an accurate and efficient solution for fall prediction using computer vision.
• The extracted features are camera-independent, reducing the need for expensive equipment and extensive modifications.
• This advancement in fall prediction technology has the potential to significantly improve safety and proactive fall prevention.