FacialOcclusionNet- A Novel Real Time Face Mask Detection Model

The COVID-19 pandemic highlighted the critical need for automated face mask detection systems to ensure compliance with health regulations in public spaces. This study proposes a FacialOcclusionNet(FONet) model for real-time face mask detection, leveraging deep learning techniques to improve accuracy and computational efficiency. The research explores conventional and deep learning-based face mask detection approaches, highlighting their limitations in real-world scenarios such as varying mask types, occlusions, and lighting conditions. The dataset consists of 7,553 RGB images, categorized into masked and unmasked faces, with data augmentation techniques applied to enhance model robustness. Two models were evaluated: a custom CNN and a fine-tuned MobileNet model. The custom CNN model achieved 96.69% accuracy, while the fine-tuned MobileNet model achieved 99.34% accuracy, demonstrating superior precision and recall. The proposed FacialOcclusionNet model, optimized for real-time inference, outperforms traditional models by balancing accuracy and computational efficiency. Results indicate that the model generalizes well under various conditions, making it suitable for deployment in high-traffic environments such as airports, railway stations, and shopping malls. Despite its high accuracy, challenges remain in detecting improperly worn masks and handling extreme lighting variations. Future work aims to enhance model adaptability through multi-modal biometric integration, and edge computing for optimized real-world applications. Unlike existing MobileNetV2-based models, our proposed FacialOcclusionNet incorporates lightweight optimizations and enhanced feature extraction that leads to higher real-time detection efficiency. This enhances AI-driven public health monitoring by reducing false positives and improving detection under occlusions and varying lighting conditions.