Real-time PPE Detection & Open Dataset (Open Access paper link)
The repository introduces eight DL models built on You Look Only Once (YOLO) architecture for PPE detection. Meanwhile, a novel high-quality dataset is constructed for detecting the person, the vest, and four helmet colors.
A novel dataset is constructed for detecting the helmet, the helmet colors and the person for this project, named Color Helmet and Vest (CHV) dataset.
Instead of just accepting exiting images, strict criteria are designed at the beginning, and only 1,330 high-quality images among 10,000 ones from the Internet and open datasets are selected.
The dataset is open for free use, please download at Google Drive or Baidu Yunpan (password: f003).
If the dataset helpes you, please cite the repository in your article:
@Article{wang2021ppe, AUTHOR = {Wang, Zijian and Wu, Yimin and Yang, Lichao and Thirunavukarasu, Arjun and Evison, Colin and Zhao, Yifan}, TITLE = {Fast Personal Protective Equipment Detection for Real Construction Sites Using Deep Learning Approaches}, JOURNAL = {Sensors}, VOLUME = {21}, YEAR = {2021}, NUMBER = {10}, ARTICLE-NUMBER = {3478}, URL = {https://www.mdpi.com/1424-8220/21/10/3478}, ISSN = {1424-8220}, DOI = {10.3390/s21103478} }
Or
Wang, Z.; Wu, Y.; Yang, L.; Thirunavukarasu, A.; Evison, C.; Zhao, Y. Fast Personal Protective Equipment Detection for Real Construction Sites Using Deep Learning Approaches. Sensors 2021, 21, 3478. https://doi.org/10.3390/s21103478
- YOLO v5x owns the best mAP, 86.55%.
- YOLO v5s has the faster processing speed, 52 FPS.
- For YOLO v3 models, different detection layers are tested, while the more layers cannot improve the performance.
- For YOLO v4 models, the increase of training image size cannot contribute to better performance.
If you are interested in my work, please visit https://zijianwang1995.github.io/