Helmet Detection System 🪖

Project Type: Individual project

Technologies Used:Python, TensorFlow, OpenCV, YOLO (You Only Look Once), Pandas, Matplotlib, Seaborn

Overview

This project aimed to develop an object detection system to identify whether individuals are wearing helmets in images. Ensuring that individuals wear helmets is crucial for safety in various industries, especially construction and transportation. The system uses the YOLO (You Only Look Once) model, known for its real-time detection capabilities, to detect helmets in images efficiently. Our goal was to create an accurate and fast system that could be implemented for safety monitoring. I was responsible for implementing and optimizing the YOLO model for helmet detection.

Key Contributions

✅ Model Development: Implemented and fine-tuned YOLO to detect helmets in real-time images.

✅ Data Preprocessing: Labeled the dataset, augmented images (through flipping, scaling, and rotation), resized them to fit the input size of YOLO, and normalized pixel values.

✅ Training & Optimization: Used transfer learning with pre-trained YOLO weights, optimized anchor boxes, and adjusted hyperparameters to improve accuracy and inference speed.

✅ Evaluation & Comparison: Evaluated the YOLO model based on accuracy, precision-recall, and inference speed, ensuring it met real-time requirements for helmet detection.

✅ Visualization: Generated bounding box visualizations on the test images and analyzed training/validation accuracy and loss over epochs.

Results

📌 YOLO Model Accuracy: 94.85% (Final model used for deployment)📌 Inference Speed: YOLO provided real-time detection, making it suitable for practical applications.📌 Confusion Matrix: The confusion matrix demonstrated strong performance in distinguishing helmet-wearing from non-helmet-wearing individuals.

🔗 Interactivity & Links

📌 GitHub Repository: (https://github.com/Yonatan8475/Helmet-Detection-.git)