SH17: A Dataset for Human Safety and Personal Protective Equipment Detection in Manufacturing Industry.

Workplace safety in manufacturing has taken a step forward thanks to a new open-source dataset designed to improve detection of personal protective equipment (PPE). The SH17 dataset, created by two researchers at the University of Windsor, contains more than 8,000 annotated images covering 17 categories of PPE and body parts, including helmets, gloves, safety glasses, and protective suits. By combining inclusivity, scale, and cutting-edge AI, SH17 offers industries a practical tool to reduce accidents, comply with safety regulations, and protect workers in high-risk environments.

“Ensuring PPE compliance is critical, but traditional monitoring methods are costly and prone to error,” said Dr. Afshin Rahimi, co-author of the study published in the Journal of Safety Science and Resilience. “By introducing SH17, we provide industries with a scalable, non-invasive solution that leverages computer vision to safeguard workers more effectively.”

The researchers trained state-of-the-art object detection models, including the latest YOLOv9 variant, which achieved a mean average precision of 70.9% in identifying PPE. “This performance not only surpasses many existing datasets but also demonstrates strong generalization when tested across different industrial environments, added Rahimi.

Co-author Hafiz Mughees Ahmad explained, “What makes SH17 unique is its diversity. We sourced images globally to minimize bias and included small, hard-to-detect items like earmuffs and tools. This ensures the dataset reflects real-world complexity rather than controlled lab conditions.”

The findings shine new light on how artificial intelligence can transform occupational safety. “Unlike earlier datasets that focused narrowly on construction helmets or vests, SH17 expands coverage to manufacturing environments, addressing a gap in existing research,” added  Ahmad.

The dataset is freely available to the public, along with trained model weights, enabling companies and researchers to build upon the work.

Looking ahead, the researchers plan to refine detection of minority classes, such as faceguards and safety vests, which had fewer samples. “Improving accuracy for these categories will make PPE monitoring even more robust,” Rahimi said.

Contact author: 

Afshin Rahimi, Associate Professor, Ph.D., P.Eng., SMIEEE, Department of Mechanical, Automotive and Materials Engineering, Rm. 2174 CEI, University of Windsor, 401 Sunset Avenue, Windsor, ON, Canada, N9B 3P4, Phone: +1 (519) 253-3000 ext. 5936 | Fax: +1 (519) 971-7007, E-mail: arahimi@uwindsor.ca

Funder: 

This study is supported by IFIVEO CANADA INC., Mitacs through IT16094, Natural Sciences and Engineering Research Council of Canada (NSERC) through ALLRP 560406-20, Ontario Center of Innovation (OCI) through OCI# 34166, and the University of Windsor, Canada.

Conflict of interest:

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

See the article:

H. M. Ahmad and A. Rahimi, “SH17: A dataset for human safety and personal protective equipment detection in manufacturing industry,” Journal of Safety Science and Resilience, vol. 6, no. 2, pp. 175–185, June 2025, doi: https://doi.org/10.1016/j.jnlssr.2024.09.002.