Research on fall detection algorithm for older adults in complex lighting environments based on improved YOLOv8n.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Gerontology

Background:

• Elderly falls pose a significant public health challenge due to global aging.
• Real-time fall detection is vital for older adult safety.
• Existing methods struggle with complex backgrounds and low accuracy.

Purpose of the Study:

• To develop an accurate and efficient fall detection method for older adults.
• To address challenges in complex lighting and shadow environments.
• To improve upon the YOLOv8n network for fall detection.

Main Methods:

• Constructed a multi-pose human fall database with diverse scenarios.
• Integrated seven advanced lightweight modules and attention mechanisms into YOLOv8n.
• Conducted 20 improvement experiments, analyzing single, dual, and triple module fusions.

Main Results:

• The dual-module enhancement (C2f_PKI and SimSPPF) proved most effective.
• Achieved 91.8% mAP@0.5 with 41.6 FPS on a self-constructed dataset, outperforming YOLOv5s and Faster-RCNN.
• Demonstrated up to 10% accuracy improvement on the UR Fall dataset in real-world scenarios.

Conclusions:

• The improved YOLOv8n model offers efficient and reliable fall detection for older adults.
• This technology has significant implications for enhancing elderly safety and promoting healthy aging.
• The study validates the model's effectiveness in complex, real-world conditions.