Helmet detection using YOLOv5

No Thumbnail Available
Authors
Pralhad Gore, Vivekanand
Issue Date
2023-08
Degree
MSc in Information Systems with Computing
Publisher
Dublin Business School
Rights
Items in eSource are protected by copyright. Previously published items are made available in accordance with the copyright policy of the publisher/copyright holder.
Abstract
Helmet detection is an essential task in computer vision and artificial intelligence, aimed at automatically identifying and localizing helmets in images or videos. The goal is to enhance safety in various domains, such as construction sites, sports events, and industrial settings, where the usage of helmets is critical for preventing head injuries. By employing advanced object detection algorithms, helmet detection systems can accurately detect the presence of helmets and draw bounding boxes around them, allowing for effective monitoring and compliance enforcement. YOLO (You Only Look Once) is a popular family of real-time object detection algorithms known for its speed and accuracy. YOLO processes images in a single pass and directly predicts bounding boxes and class probabilities without using any region proposal methods. YOLOv5 is one of the latest iterations in the YOLO series, boasting improved performance and an efficient architecture, making it an ideal choice for real-time object detection tasks. In this research project, we chose to employ YOLOv5 for helmet detection due to its impressive real-time processing capabilities and accurate detection results. Additionally, YOLOv5 is relatively lightweight compared to its predecessors, making it more accessible for integration into web applications and systems with limited computational resources. To begin our work, we trained the YOLOv5 model using the popular Google Colab platform, utilizing its GPU resources for faster training. The training data consisted of annotated images of helmets, and we fine-tuned the model to ensure it learned the intricate features of helmets accurately. Upon successful training, we proceeded to integrate the trained YOLOv5 model into a user-friendly Flask web application. The application enabled users to upload images and videos, which were then processed by the model to detect helmets. The detected images and videos were displayed back to users with bounding boxes around the identified helmets, providing a visual representation of the detection results. Throughout the process, we meticulously tuned the model's parameters to strike a balance between detection accuracy and processing speed. This was crucial to ensure that the system performed efficiently while accurately identifying helmets in real-world scenarios. Our research project successfully developed a Helmet Detection system using YOLOv5, offering a practical solution to enhance safety measures by automating helmet identification and monitoring in various contexts. The combination of YOLOv5's speed and accuracy, along with the user-friendly web application, makes our system a valuable tool for safety compliance and injury prevention in helmet-required environments.