SHEL5K: An Extended Dataset and Benchmarking for Safety Helmet Detection

Area of Science:

• Computer Vision
• Deep Learning
• Industrial Safety

Background:

• Automatic helmet detection is crucial for safety in construction and manufacturing.
• Existing public datasets lack sufficient labeled data and classes for robust deep learning models.
• Developing accurate helmet detection models requires large, well-annotated datasets.

Purpose of the Study:

• To introduce the Safety HELmet dataset with 5K images (SHEL5K), an improved version of the SHD dataset.
• To provide a comprehensive dataset with six fully labeled classes for enhanced helmet detection.
• To evaluate the performance of various state-of-the-art object detection models on the new dataset.

Main Methods:

• The study presents the SHEL5K dataset, featuring 5,000 images with six detailed labels: helmet, head, head with helmet, person with helmet, person without helmet, and face.
• Multiple object detection models, including YOLOv3, YOLOv4, YOLOv5, Faster R-CNN, and YOLOR, were tested using the SHEL5K dataset.
• Performance was evaluated based on mean Average Precision (mAP) to compare model effectiveness.

Main Results:

• The SHEL5K dataset demonstrated superior performance compared to existing datasets due to its comprehensive labeling and increased number of classes.
• Tested object detection models showed improved mean Average Precision (mAP) when trained and evaluated on the SHEL5K dataset.
• The enhanced dataset facilitates more accurate and reliable automatic helmet detection.

Conclusions:

• The SHEL5K dataset is a valuable resource for advancing research in automatic safety helmet detection.
• The dataset's quality and class diversity contribute to achieving higher accuracy in identifying helmet compliance.
• This work supports the development of more effective safety monitoring systems in high-risk industries.