Real-Time Detection of Personal Protective Equipment in Video Streams Using YOLOv8 Deep Learning
Midocean Journal for Research and Studies · 12.06.2026 · 2 (3) · 312-330
DOI: 10.71335/g2q35w59 ↗Kurz gefasst
Fachartikel in der Zeitschrift Midocean Journal for Research and Studies, online erschienen am 12.06.2026. Verfasst von Younis Arrabi.
Abstract (Originalsprache)
Objectives
This study develops and evaluates a real-time automated system for detecting the compliance of Personal Protective Equipment (PPE) in industrial video streams in order to solve the main bottleneck existing between the manual inspection practices and the necessity of continuous and scalable safety monitoring. Methodology: Experimental methodology was incorporated. A custom dataset of 1,400 annotated instances of PPE was created and expanded to around 9,000 images using geometric transformations, greyscale conversion and the addition of Gaussian noise. Five classes of PPE were targeted: Helmets, Safety vests, gloves, safety shoes and safety goggles. The dataset was divided into training, validation, and test sets with 70%, 20%, and 10% of the data, respectively. YOLOv8 was trained using GPU accelerated platform (Google Colab, Nvidia fast) 50 epochs with Convolutional Neural Network (CNN) benchmarking (mAP, Precision, Recall & F1-Score).
Results
YOLOv8 achieved mAP of 98.4%, precision 98.2% and recall 98.2% and F1-Score = 98.3% whereas CNN had a poor result of 95.9%. YOLOv8 also outperforms the CNN baseline and all the comparable previous studies in 22 detectors corresponding to the multi-class PPE detection problem. Results by per-class were highest for helmets (mAP = 99.5%, F1 = 99.95%) and safety vests (mAP = 99.5%). Six live detection tests proved the ability of real time inference. Recommendations: The system is recommended to be embedded as part of the existing CCTVs in construction and industrial environments. Future work has encountered the need to expand the classification for safety goggles, study the compression of edge device model as well as apply transformer-based structures for detecting small objects better.
Zitieren
APA
Arrabi, Y. (2026). Real-Time Detection of Personal Protective Equipment in Video Streams Using YOLOv8 Deep Learning. Midocean Journal for Research and Studies, 2(3), 312-330. https://doi.org/10.71335/g2q35w59
Vancouver
Arrabi Y. Real-Time Detection of Personal Protective Equipment in Video Streams Using YOLOv8 Deep Learning. Midocean Journal for Research and Studies. 2026;2(3):312-330. doi:10.71335/g2q35w59
BibTeX
@article{Arrabi2026,
author = {Arrabi, Younis},
title = {Real-Time Detection of Personal Protective Equipment in Video Streams Using YOLOv8 Deep Learning},
journal = {Midocean Journal for Research and Studies},
year = {2026},
volume = {2},
number = {3},
pages = {312--330},
doi = {10.71335/g2q35w59}
}
EUVIMED