Workplace Safety and Access Control Using YOLO for Helmet and Face Detection

Abstract

The impact of a dangerous workplace on employee health and productivity has caused many firms to place a high priority on workplace safety. Employees are constantly exposed to a variety of risks at all times and places when working in today\'s vast construction/manufacturing complexes and other dangerous industrial locations. The incidence of accidents is therefore higher than in other industries due to the greater number of risk variables, and it is mandatory that workers wear personal protective equipment (PPE) to shield their bodies from hazardous causes. Accidents resulting from failure to wear personal protection equipment, such as safety helmets, are among the most frequent safety incidents at industrial sites. In actuality, the majority of current safety inspection procedures depend on the manual monitoring and reporting of inspectors. Observing construction sites by hand can be labour-intensive, prone to mistakes, expensive, and unsuitable for larger projects with multiple ongoing operations. Many studies on automatic helmet wearing detection and human identity identification have been published, with the goal of assisting safety inspectors on construction sites in their duties of monitoring worker safety. Another study claims that helmet wear can be integrated with person identification using computer vision. Stated differently, during helmet testing, we typically lack the ability to identify specific individuals, and vice versa. We suggest a computer vision-based approach to automatically detect workers\' identity and helmet wear in order to address the aforementioned issues. Firstly, our approach combines two uses: identification and helmet wearing detection. Second, we evaluated the algorithm\'s accuracy and recall rate in various visual settings to determine its applicability in the actual construction site environment. This was done in accordance with the varying visual conditions on the construction site.

Introduction

I. Worker Safety in Construction

Worker safety on construction sites is critical due to hazardous conditions. Key measures include:

• Comprehensive safety training and regular refresher sessions.

• Personal Protective Equipment (PPE) like helmets, gloves, and safety shoes.

• Fall protection systems, routine machinery maintenance, and proper usage training.

• Effective hazard communication through signage and meetings.

• Safety issue documentation to promote a proactive safety culture.

Helmets play a vital role in protecting against head injuries and comply with safety regulations. They consist of a hard shell and suspension system, and come in various types suited to different job roles.

II. Related Work (Helmet Detection Systems)

Several research works have improved helmet detection using AI and computer vision:

1. Desu Fu et al.:

   • Improved YOLOv5 for small target detection.

   • Accuracy improved to 95%, with helmet recognition at 94.6%.

   • Valuable for real-time detection in power industry environments.

2. Shuai Wang et al.:

   • Used a combination of GMM, MHOG+SVM, OpenPose, and CNN.

   • Achieved 99.43% accuracy even under varied lighting and occlusions.

3. Yi-Jia Zhang et al.:

   • Developed a color and contour-based detection algorithm.

   • Works well for batch image detection, though lacks real-time capability.

4. Ahatsham Hayat et al.:

   • Developed a YOLOv5x-based system using a 5,000-image dataset.

   • Achieved 92.44% mAP, effective even in low-light settings.

5. Han Liang et al.:

   • Proposed a lightweight model using GhostNet and MSFFN for multiscale detection.

   • Their LRCA-Netv2 achieved 93.5% mAP and 42 FPS, improving both speed and accuracy.

III. Existing Methodologies

• Traditional manual video monitoring is time-consuming and less accurate.

• Sliding-window object detection methods are inefficient and may miss helmets due to varying sizes and angles.

• Shift towards automated computer vision and machine learning systems for real-time, accurate helmet detection is gaining traction.

IV. Proposed Methodology

A combined facial recognition and helmet detection system is proposed:

• Grassmann Algorithm for facial recognition:

  • Extracts facial features using image registration, tangent space mapping, and subspace distance computation.

• YOLO Algorithm for helmet detection:

  • Detects objects in real-time via bounding box predictions and class scores.

• Real-time monitoring and alert system:

  • Triggers alarms for non-compliance (no helmet or unauthorized personnel).

  • Aims to improve workplace safety through timely interventions.

V. Results & Discussion

• System testing was conducted using a standard facial recognition database.

• Highlights the importance of benchmark datasets for fair algorithm evaluation.

• The integrated system demonstrates strong potential for improving safety compliance in real-world settings.

Summary Statement

This paper emphasizes the importance of construction site safety, particularly helmet usage, and presents a hybrid AI-based solution that combines facial recognition (Grassmann algorithm) and real-time helmet detection (YOLO). Prior work and proposed methods all aim to automate and enhance safety compliance through accurate, real-time monitoring systems.

Conclusion

Using the Grassmann and YOLO algorithms, the suggested system combines helmet detection and facial recognition, offering a sophisticated outcome. With the help of real-time data processing and well-placed cameras, this system not only enforces safety regulations by looking for safety helmets but also guarantees allowed entry through facial recognition. The YOLO algorithm efficiently streamlines helmet detection, while the Grassmann method makes exact facial feature extraction possible, offering a dependable foundation for identification. By combining these cutting-edge technologies, a comprehensive approach to workplace safety is provided, enabling prompt reactions to instances of unauthorized entry and non-compliance with helmet regulations via proactive alarm systems. With its strong and flexible approach to risk mitigation, the suggested system is at the cutting edge of innovation in the construction industry, which is still very focused on safety.

References

[1] Cheng, Rao, et al. \"Multi-scale safety helmet detection based on SAS-YOLOv3-tiny.\" Applied Sciences 11.8 (2021): 3652. [2] Deng, Lixia, et al. \"A lightweight YOLOv3 algorithm used for safety helmet detection.\" Scientific reports 12.1 (2022): 10981. [3] Fu, Desu, et al. \"Research on safety helmet detection algorithm of power workers based on improved yolov5.\" Journal of Physics: Conference Series. Vol. 2171. No. 1. IOP Publishing, 2022. [4] Hayat, Ahatsham, and Fernando Morgado- Dias. \"Deep learning-based automatic safety helmet detection system for construction safety.\" Applied Sciences 12.16 (2022): 8268. [5] Hayat, Ahatsham, and Fernando Morgado- Dias. \"Deep learning-based automatic safety helmet detection system for construction safety.\" Applied Sciences 12.16 (2022): 8268. [6] Huang, Li, et al. \"Detection algorithm of safety helmet wearing based on deep learning.\" Concurrency and Computation: Practice and Experience 33.13 (2021): e6234. [7] Liang, Han, and Suyoung Seo. \"Automatic detection of construction workers’ helmet wear based on lightweight deep learning.\" Applied Sciences 12.20 (2022): 10369 [8] Li, Yange, et al. \"Deep learning-based safety helmet detection in engineering management based oconvolutional neural networks.\" Advances in Civil Engineering 2020 (2020): 1-10 [9] Wang, Shuai, et al. \"An intelligent vision- based method of worker identification for industrial internet of things (IoT).\" Wireless Communications and Mobile Computing 2022 (2022). [10] Zhang, Yi-Jia, Fu-Su Xiao, and Zhe-Ming Lu. \"Safety Helmet Wearing Detection Based on Contour and Color Features.\" Journal of Network Intelligence: Taiwan, China 7 (2022): 516-525.

Copyright

Copyright © 2025 G. Nithiyabharathi, A. Ananthi, M. Mohamed Faisal. 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.