Real-Time Personal Protective Equipment Compliance Monitoring Using Deep Learning Techniques

Abstract

This paper, we provide a real-time detection approach that detects Personal Protective Equipment using YOLOv8, a Latest object detection model.Itisusedtokeepaneyeonconstructionsites or industrial environments and confirm whether workers are using appropriate protective gear like hardhatsandsafetyvests.Themodelreceivesinput from a webcam or image dataset runs a Personal Protective Equipment (PPE) violation detection model on the video stream or images, and sends alertsinrealtimeonTelegram. Ourimplementation uses Python, OpenCV (Open-Source Computer Vision Library), and the Ultralytics YOLOv8 implementation. This project presents a viable safety compliance monitor that can easily be integrated intoalargersystemFortheProtectionandsafetyof the workers.

Introduction

Personal Protective Equipment (PPE) such as hard hats, safety vests, and masks are vital for preventing injuries in high-risk workplaces like construction sites. However, manual monitoring of PPE compliance is often inefficient, inconsistent, and prone to human error, especially in large or dynamic environments.

This project proposes a real-time PPE compliance monitoring system using YOLOv8, a state-of-the-art deep learning object detection model, integrated with Telegram alerts for instant notifications. The system analyzes live video feeds to detect PPE violations quickly (under 100 ms per frame) and alerts supervisors immediately to address hazards, enhancing workplace safety.

The project uses a diverse and well-annotated dataset of over 2,800 images and video from actual construction sites, covering various lighting and environmental conditions. The YOLOv8 model, trained and validated on this dataset, achieves high accuracy (around 90%) in detecting PPE items like helmets, vests, and masks. The system features automated Telegram alerts with attached evidence of violations, a real-time compliance dashboard, and a cooldown mechanism to prevent alert spam.

Literature review shows a progression from earlier YOLO versions to YOLOv8, with improvements in speed, accuracy, and real-time capabilities. Integration with IoT and messaging platforms enhances proactive safety management. Challenges include low-light detection, computational resource demands, scalability for large sites, and adapting to real-world variability.

Overall, the system automates PPE monitoring, reduces human error, and helps enforce safety protocols promptly, aiming to reduce workplace accidents and promote a safer working culture.

Conclusion

The evaluation findings of the YOLOv8 object detectionmodelonthePPEdetectiondemonstrated meaningful evidence of the model\'s effectiveness and potential for real-world applications. The sustained increase in accuracy throughout its training denotes the increasing accuracy of the model to detect PPE, emphasizing its ability to developareliabletoolinidentifying eventsofnon- complianceand safetyviolations. Theabilityofthe YOLOv8 object detection model to differentiate between the presence of PPE from the manner of misuse,suggeststhatthemodelcanproduceareliable recommendation regarding a specific intervention to address misuse. The sustained improvement in recall performance acrosstrainingepochsfurtherincreasescredenceto the models increasing experience tracking PPE- related behaviors over time. These findings further strengthen the ability of the YOLOv8 object detection model to support safety compliance in industrial and construction contexts. Overall, these findings describe the evolution of the YOLOv8 modelfromloweraccuracyandreliabilitytohigher accuracy and reliability and the development of a useful basis from which to make practical and pertinent recommendations regarding PPE. The successful application of the YOLOv8 object detection model supports its generalizability and application to real-world contexts.

References

[1] A study by Redmon and Farhadi titled \"YOLOv3:Anincrementalimprovement\"discusses advancements made in the YOLOv3 algorithm. [2] Redmon, Divvala, Girshick, and Farhadi present the original YOLO algorithm in their paper \"You onlylookonce:Unified,real-timeobjectdetection\" at the IEEE conference on computer vision and pattern recognition in 2016. [3] Redmon and Farhadi propose YOLO9000, an improved version of YOLO, in their paper \"YOLO9000: better, faster, stronger.\" [4] Li, Li, Li, and Cui introduce a PPE detection system based on deep learning in their article published in Sensors journal. [5] Chen,Zhang,andWangpresentareal-timePPE detection system using YOLOv3 in the Journal of Physics: Conference Series. [6] Zhang, Yang, Fu, Zhao, Wang, and Ye discuss theautomaticdetectionofPPEinconstructionsites basedontheFasterR-CNNalgorithmintheJournal of Computing in Civil Engineering. [7] Jin, Jiang, Sun, and Cao propose an improved Faster R-CNN algorithm for PPE detection in construction sites in the Applied Sciences journal. [8] Ren,He,Girshick,andSunintroducetheFaster R-CNN algorithm for real-time object detection in their paper published in IEEE Transactions on Pattern Analysis and Machine Intelligence. [9] Shang, Mao, Cheng, Zhao, and Zhou present a novel deep learning-based method for detecting PPE in the Journal of Computational Science. [10] Zhang, Yang,Tian,Wang, and Zhangdiscuss PPEdetectionbasedondeeplearningintheirpaper presented at the International Conference on Cyber Security and Privacy Engineering. [11] Kim,Lee,andKimproposeFastMaskR-CNN for PPE detection in the Applied Sciences journal. [12] Wei, Fan, Yu, and Zhao present an object detection model based on YOLOv5 for PPE in the construction industry in Engineering Reports. [13] Kuo, Lu, and Chou discuss detecting PPE using Mask R-CNN for occupational safety in the Sensors journal. [14] Bochkovskiy,A.,Wang,C. Y.,&Liao, H.Y.M.(2020).YOLOv4:OptimalSpeedand AccuracyofObject Detection.arXivpreprint arXiv:2004.10934. [15] Liu,W.,Anguelov,D.,Erhan,D.,Szegedy,C.,Reed,S.,Fu,C.Y.,&Berg,A.C.(2016).SSD:SingleShot MultiBox Detector. European Conference on Computer Vision (ECCV). [16] Ren,S.,He,K.,Girshick,R.,&Sun,J.(2015).Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. Advances in Neural Information Processing Systems (NeurIPS). [17] Jocher,G.,etal.(2021).YOLOv5by Ultralytics. GitHub Repository. [18] Wang, C.Y., Bochkovskiy, A.,&Liao, H. Y.M. (2022). YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real-Time Object Detectors. arXiv preprint arXiv:2207.02696. [19] Li,Z.,etal.(2022).YOLOv6:ASingle-Stage Object Detection Framework for Industrial Applications. arXiv preprint arXiv:2209.02976. [20] Ultralytics.(2023).YOLOv8:State-of-the-Art Object Detection. GitHub Repository. [21] Medini. (2020). Personal Protective Equipment (PPE) DetectionUsingDeepLearning Techniques. Medium. [22] MDPI Safety. (2023). Deep Learning for Detection of Proper Utilization and Adequacy of Personal Protective Equipment. MDPI Safety. [23] Sandru, A., Duta, G. E., Georgescu, M. I., & Ionescu, R. T. (2020). SuPEr-SAM: Using the SupervisionSignalfromaPoseEstimatortoTraina Spatial Attention Module for Personal Protective Equipment Recognition. arXiv preprint arXiv:2009.12339. [24] Cord. (2023). YOLOv8 for Object Detection Explained [Practical Example]. Medium. [25] Nature. (2022). The Compliance of Head- Mounted Industrial PPE by Using Deep Learning. Scientific Reports. [26] DigitalOcean. (2023). Object Detection with YOLOv8 Advanced Capabilities. DigitalOcean Community Tutorials.

Copyright

Copyright © 2025 Geetha G, Diwakar S, Karthikeyan A S, Ragul M N, Roopesh N. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.