Helmet Wearing State Detection Based on Improved Yolov5s

Area of Science:

• Computer Vision
• Artificial Intelligence
• Construction Safety Engineering

Background:

• Helmet use is critical for worker safety in construction.
• Existing detection algorithms lack nuanced classification of helmet-wearing states.
• There's a need for advanced monitoring tools in construction safety.

Purpose of the Study:

• To develop a more comprehensive helmet-wearing detection system for construction sites.
• To improve upon existing helmet detection algorithms by introducing finer classifications.
• To enhance the accuracy and performance of helmet detection models.

Main Methods:

• A specialized dataset with six distinct helmet-wearing scenarios was created.
• The YOLOv5s algorithm was enhanced with K-means for prior box optimization.
• An additional layer was added for improved small target recognition.
• Attention mechanisms and EIOU_Loss function were integrated into the model.

Main Results:

• The improved YOLOv5s algorithm demonstrated higher accuracy compared to the original version.
• Finer classification of helmet-wearing states significantly boosted detection performance.
• The algorithm effectively distinguished between correct and incorrect helmet usage.

Conclusions:

• The proposed enhanced YOLOv5s algorithm offers superior performance for construction helmet detection.
• This advanced detection system can significantly contribute to improving construction site safety.
• The methodology provides a robust framework for nuanced object detection in real-world applications.