CNN-based helmet compliance detector using VGG16, achieving 99.6% accuracy for automated safety monitoring.
This project presents a deep learning solution for automatically detecting helmet compliance among workers in hazardous environments such as construction sites and industrial settings. The system uses a CNN-based binary image classifier to identify whether a person is wearing a safety helmet, helping improve safety enforcement by replacing manual monitoring with an accurate, automated vision-based approach.
Model: Built a Convolutional Neural Network (CNN) using transfer learning with VGG16 to classify images into two categories: helmet and no helmet. Performance: Achieved 99.6% accuracy, recall, precision, and F1-score on test data. Automation: Provides a scalable computer vision system for real-time or batch safety compliance monitoring. Data Pipeline: Used ImageDataGenerator for training and validation splits, along with data augmentation to improve model robustness.
Deep Learning: TensorFlow, Keras Computer Vision: OpenCV, VGG16 Model Architecture: CNN, Feed Forward Neural Network Data Handling: ImageDataGenerator, Data Augmentation