Ensuring workplace safety is a crucial challenge across industries. Personal Protective Equipment (PPE) compliance plays a significant role in mitigating risks, yet manual monitoring can be error-prone and resource-intensive. This project presents a computer vision-based PPE detection system designed to automate the identification of compliance and non-compliance in real-time. Leveraging state-of-the-art machine learning models and a robust dataset, this solution demonstrates significant potential for enhancing safety protocols in dynamic environments.
Workplace accidents, often resulting from negligence in wearing mandatory safety gear, remain a pressing concern in construction and manufacturing sectors. Traditional manual inspections, though effective, are limited by human fatigue and scalability issues. This project aims to address these challenges by developing an automated PPE detection system capable of identifying essential safety gear such as helmets, safety vests, and shoes in real-time. By utilizing computer vision and deep learning technologies, the system ensures enhanced monitoring accuracy and efficiency.
The project utilizes the PPE Detection dataset from Roboflow, accessible at Roboflow PPE Detection Dataset, comprising:
2197 images and Five distinct classes: Helmet, No Helmet, Safety Vest, No Safety Vest, Shoes.
Preprocessing involved resizing images to 640x640 pixels and applying augmentations like:
Frameworks and Tools: The model was developed using the YOLOv5s architecture, renowned for its
balance of speed and accuracy in object detection tasks.
Training Details:
Real-Time Detection: Integrated the model into a live system using Streamlit.
Deployment Setup: Connected the model to video streams to monitor PPE compliance in real-world
environments.
Performance Metrics
Accuracy: 92.5%
Precision: 90.2%
Recall: 93.8%
mAP (Mean Average Precision): 91.7%
Output Examples
Below is an example of the detection output highlighting violations and compliance:
- Image 1: Workers wearing helmets and vests marked as compliant.
- Image 2: Workers without helmets flagged for non-compliance.
Annotated Images:
Include images generated from the model’s detection outputs with bounding boxes and labels.
This project underscores the potential of computer vision in revolutionizing workplace safety monitoring. By automating PPE compliance checks, organizations can achieve higher safety standards while optimizing resource utilization. The system’s adaptability ensures its relevance across a wide range of industries.