Lightweight Attention-Augmented YOLOv5s for Accurate and Real-Time Fall Detection in Elderly Care Environments

Area of Science:

• Computer Vision
• Artificial Intelligence
• Geriatric Care Technology

Background:

• Elderly falls are a major global health concern, leading to significant injury and mortality.
• Existing monitoring systems often lack the accuracy, efficiency, or real-time capabilities required for effective intervention.
• There is a critical need for advanced, non-intrusive solutions to ensure the safety of the aging population.

Purpose of the Study:

• To develop a lightweight, accurate, and efficient AI framework for real-time elderly fall detection.
• To enhance the YOLOv5s model with attention mechanisms and optimized feature fusion for improved performance.
• To create a robust system deployable in real-world elderly care settings.

Main Methods:

• An improved YOLOv5s architecture was developed, incorporating a Convolutional Block Attention Module (CBAM) for salient feature enhancement.
• Multi-scale feature fusion in the Neck was optimized for better detection of small objects, and anchor boxes were re-clustered for fall morphology.
• A diverse dataset of 11,314 images across multiple scenes was used for training and evaluation.

Main Results:

• The improved YOLOv5s model achieved a mean average precision (mAP@0.5) of 94.2% and a recall of 92.5%.
• The system demonstrated a low false alarm rate of 4.2% and maintained real-time detection speeds at 32 frames per second (FPS).
• Performance surpassed baseline YOLOv5s and YOLOv4 models, indicating enhanced robustness and generalization.

Conclusions:

• The integration of attention mechanisms, adaptive fusion, and anchor optimization significantly improves fall detection accuracy and reliability.
• The developed framework offers a scalable and deployable AI solution for intelligent, non-intrusive safety monitoring in aging populations.
• Future work may involve multimodal fusion and illumination-invariant modeling to address performance limitations in extreme conditions.