SDES-YOLO: A high-precision and lightweight model for fall detection in complex environments

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biomedical Engineering

Background:

• Falls pose significant health risks, necessitating rapid detection for timely intervention.
• Existing object detection models like YOLOv8 struggle with fall detection due to factors like lighting, occlusion, and posture variations.
• Accurate and efficient fall detection is critical for improving patient safety and quality of life.

Purpose of the Study:

• To develop an improved fall detection model, SDES-YOLO, based on YOLOv8.
• To enhance the model's ability to handle complex scenarios including occlusions and varied postures.
• To achieve high accuracy and computational efficiency for practical fall detection applications.

Main Methods:

• Proposed the SDES-YOLO model, an enhancement of YOLOv8.
• Integrated a multi-scale feature extraction pyramid (SDFP) for comprehensive feature capture.
• Incorporated an occlusion-aware attention mechanism (SEAM) and an edge-spatial information fusion module (ES3).
• Utilized a WIoU-Shape loss function to optimize detection performance.

Main Results:

• SDES-YOLO achieved an mAP@0.5 of 85.1%, a 3.41% improvement over YOLOv8n.
• The model demonstrates high efficiency with only 2.9M parameters and 7.2 GFLOPs.
• Achieved a reduction in parameter count (1.33%) and computation (11.11%) compared to YOLOv8n.
• Successfully improved fall detection performance in complex and challenging scenarios.

Conclusions:

• SDES-YOLO effectively enhances fall detection accuracy and precision.
• The model offers optimized computational efficiency, making it suitable for resource-constrained environments.
• SDES-YOLO represents a significant advancement in AI-powered fall detection systems for improved safety.