SH17
The SH17 dataset was created by the Department of Mechanical, Automotive and Materials Engineering at the University of Windsor, containing 8,099 annotated images covering 17 categories of personal protective equipment (PPE) such as helmets and safety glasses. Collected from diverse industrial settings, the dataset aims to enhance worker safety in manufacturing through object detection and convolutional neural network techniques. Images were sourced via the Pexels website and annotated by professionals to ensure quality and diversity. The dataset is primarily used to train and validate object detection models for PPE compliance in industrial environments.
Dataset description and usage context
SH17: A Dataset for Human Safety and Personal Protective Equipment Detection in Manufacturing Industry
Dataset Overview
Abstract
Workplace accidents pose significant risks to human safety, especially in construction and manufacturing sectors where PPE compliance is increasingly essential. This study leverages object detection (OD) and convolutional neural network (CNN) techniques to develop non‑intrusive technologies for detecting and verifying the correct use of various PPE (e.g., helmets, safety glasses, masks, and protective clothing). We introduce the SH17 dataset, comprising 8,099 annotated images with 75,994 instances across 17 PPE categories and body parts, collected from diverse industrial environments. We trained state‑of‑the‑art OD models for benchmarking; preliminary results show the YOLOv9‑e variant exceeding 70.9% accuracy in PPE detection. Cross‑dataset validation indicates that integrating these technologies can markedly improve safety management systems, offering scalable and efficient solutions for industries striving to meet safety regulations and protect their workforce.
Dataset Details
- Image Count: 8,099 annotated images
- Instance Count: 75,994 object instances
- Category Count: 17 PPE items and body parts
Key Features
- Collected from globally diverse industrial environments
- High‑quality images (max resolution 8192×5462, min 1920×1002)
- Average of 9.38 instances per image
- Includes small objects such as ears and earmuffs (39,764 annotations < 1% image area, 59,025 annotations < 5% area)
Categories
- Person
- Head
- Face
- Glasses
- Face‑mask‑medical
- Face‑guard
- Ear
- Earmuffs
- Hands
- Gloves
- Foot
- Shoes
- Safety‑vest
- Tools
- Helmet
- Medical‑suit
- Safety‑suit
Models
We used the ultralytics==8.0.38 repository to train all YOLO models. Training weights for YOLO v8, v9, and v10 object detection models are provided for evaluation and benchmarking.
YOLO v8, v9, v10 Results
| Model | Params (M) | Images | Instances | Precision (P) | Recall (R) | mAP@50 | mAP@50‑95 |
|---|---|---|---|---|---|---|---|
| Yolo‑8‑n | 3.2 | 1620 | 15358 | 67.5 | 53.6 | 58.0 | 36.6 |
| Yolo‑8‑s | 11.2 | 1620 | 15358 | 81.5 | 55.7 | 63.7 | 41.7 |
| … (remaining rows omitted for brevity) |
Usage
from ultralytics import YOLO
model = YOLO(r"path\to\weight.pt") # Provide path to trained model
results = model("path/to/image.jpg") # Run inference
results.show() # Visualize results
Refer to the ultralytics documentation for further details.
Evaluation
Evaluation code is provided to reproduce our benchmark results on the SH17 validation set.
from ultralytics import YOLO
model = YOLO(r"path\to\weight.pt")
model.val(data="sh17.yaml", batch=1, imgsz=640, device="cuda:0")
Citation
If you use this dataset or code in your research, please cite our paper:
@article{ahmad_2024_sh17,
title={SH17: A Dataset for Human Safety and Personal Protective Equipment Detection in Manufacturing Industry},
author={Ahmad, Hafiz Mughees and Rahimi, Afshin},
journal={arXiv},
year={2024}
}
License
The SH17 dataset is released under the CC BY‑NC‑SA 4.0 license.
Pair the dataset with AI analysis and content workflows.
Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.