Workshop Safety Helmet Wearing Detection Model Based on SCM-YOLO

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Object detection algorithms like YOLOv4-tiny face challenges in complex scenes, including inadequate feature extraction, reduced accuracy, and low recall rates.
• Accurate detection of safety helmet wearing is crucial for workplace safety compliance and accident prevention.

Purpose of the Study:

• To propose an improved YOLOv4-tiny algorithm, named SCM-YOLO, for enhanced safety helmet detection in complex environments.
• To address the limitations of existing algorithms by improving feature extraction, accuracy, and recall rates.

Main Methods:

• Integration of Spatial Pyramid Pooling (SPP) after the backbone network to enhance multi-scale feature adaptability.
• Incorporation of Convolutional Block Attention Module (CBAM), Mish activation, K-Means++ clustering, label smoothing, and Mosaic data augmentation.
• Utilizing YOLOv4-tiny architecture as the base for the proposed SCM-YOLO model.

Main Results:

• The SCM-YOLO algorithm achieved a mean Average Precision (mAP) of 93.19%, outperforming the standard YOLOv4-tiny by 4.76%.
• The improved model demonstrated an inference speed of 22.9 Frames Per Second (FPS) on a GeForce GTX 1050Ti.
• The enhancements successfully improved the detection accuracy of small objects without compromising detection speed.

Conclusions:

• SCM-YOLO offers a significant improvement over YOLOv4-tiny for safety helmet detection in complex scenarios.
• The proposed algorithm meets the requirements for real-time and accurate safety helmet detection, contributing to improved safety management.
• The combination of architectural modifications and training techniques effectively enhances object detection performance.