BMR-YOLO: A deep learning approach for fall detection in complex environments.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biomedical Engineering

Background:

• Traditional fall detection systems struggle with environmental challenges like occlusion and poor lighting.
• Real-time intelligent monitoring requires robust and accurate fall detection methods.

Purpose of the Study:

• To propose an optimized BMR-YOLO framework based on YOLOv8n for enhanced fall detection.
• To address limitations of existing methods in complex environments, particularly occlusion and lighting variations.

Main Methods:

• Enhanced backbone with BiFormer vision transformer and dual-layer routing attention.
• Replaced C2f module with C2f_rvb for improved multi-scale feature handling.
• Incorporated MultiSEAM attention mechanism and direction-aware SIoU loss for better accuracy and localization.

Main Results:

• BMR-YOLO achieved a mean Average Precision (mAP@0.5) of 0.899 on the BMR-fall dataset, an improvement from 0.852.
• Maintained low computational cost at 6.5 GFLOPs.
• Outperformed existing methods in occlusion and lighting variation scenarios.

Conclusions:

• The proposed BMR-YOLO framework demonstrates superior performance and robustness in challenging fall detection scenarios.
• The optimized architecture offers practical applicability for real-world intelligent monitoring systems.
• The study validates the effectiveness of the proposed enhancements for accurate and stable fall detection.