Advancing construction safety: YOLOv8-CGS helmet detection model

Area of Science:

• Computer Vision
• Artificial Intelligence
• Construction Safety Engineering

Background:

• Real-time object detection is vital for construction site safety, specifically for identifying safety helmets.
• Traditional methods struggle with complex construction environments like low light, occlusion, and varied helmet shapes.

Purpose of the Study:

• To develop an enhanced helmet detection model (YOLOv8-CGS) for improved accuracy and robustness in construction safety management.
• To address limitations of existing object detection techniques in challenging real-world scenarios.

Main Methods:

• Utilized the YOLOv8 architecture as a base.
• Integrated attention mechanisms: Convolutional Block Attention Module (CBAM) and Global Attention Mechanism (GAM).
• Incorporated Smooth Labeling Loss Function (SLOU) for optimized bounding box predictions.

Main Results:

• YOLOv8-CGS achieved 94.58% accuracy on the SHD dataset and 92.38% on the SHWD dataset.
• Demonstrated significant improvements of 5.9% and 5.94% over the standard YOLOv8 model.
• Showcased enhanced performance in detecting helmets under occlusion and in low-light conditions.

Conclusions:

• YOLOv8-CGS offers more efficient and accurate helmet detection for practical construction safety applications.
• The model significantly enhances real-time monitoring capabilities, contributing to improved worker safety.
• The integration of attention modules and SLOU effectively tackles challenges in complex construction environments.