Safety Helmet Detection Based on YOLOv5 Driven by Super-Resolution Reconstruction

Area of Science:

• Computer Vision
• Construction Safety Engineering
• Image Processing

Background:

• High-resolution image transmission is critical for accurate safety helmet detection in construction.
• Existing methods struggle with high-speed detection due to resolution limitations.
• Ensuring worker safety compliance requires reliable detection systems.

Purpose of the Study:

• To develop an efficient super-resolution (SR) reconstruction module for improving image resolution prior to detection.
• To enhance the speed and accuracy of safety helmet detection in the construction industry.
• To address the challenges posed by low-resolution images in construction safety monitoring.

Main Methods:

• A novel super-resolution (SR) reconstruction module incorporating a multichannel attention mechanism.
• Integration of a modified Cross Stage Partial (CSP) module within the YOLO (You Only Look Once) v5 architecture.
• Experimental validation of the proposed algorithm's performance.

Main Results:

• The proposed SR module significantly improves image quality before detection.
• Achieved a peak signal-to-noise ratio (PSNR) of 29.420.
• Attained a structural similarity index measure (SSIM) of 0.855, indicating high fidelity.

Conclusions:

• The developed algorithm effectively enhances safety helmet detection in construction environments.
• The combination of SR and attention mechanisms improves feature capture and reduces information loss.
• The model demonstrates robust performance for real-world construction safety applications.