Improving the accuracy of existing camera based fall detection algorithms through late fusion

Area of Science:

• Gerontology
• Computer Science
• Biomedical Engineering

Background:

• Falls pose a significant health risk to the elderly population.
• Existing fall detection systems often lack accuracy in real-world scenarios.
• Timely aid following a fall is crucial for mitigating consequences.

Purpose of the Study:

• To improve the accuracy of existing fall detection systems.
• To introduce a novel late fusion technique for enhanced fall detection.
• To evaluate different aggregation methods for combining confidence levels.

Main Methods:

• A late fusion technique was developed to combine confidence levels from multiple single-camera fall detection systems.
• Four distinct aggregation methods were compared.
• Performance was evaluated using the Area Under the Curve (AUC) of precision-recall curves.

Main Results:

• The proposed late fusion technique significantly improved fall detection accuracy.
• Using the median of confidence levels from five cameras resulted in a 218% increase in AUC.
• This method offers a straightforward way to enhance existing fall detection systems.

Conclusions:

• Late fusion is an effective strategy for improving fall detection system accuracy.
• Multi-camera integration via confidence level aggregation offers substantial performance gains.
• The proposed method provides a practical and easy-to-implement solution for better fall detection in older adults.