Face mask detection using deep hybrid network architectures

Abstract

As the world has been severely affected by Novel Coronavirus, scientists have been working hard to study this rapidly evolving virus, its long-term and short-term implications, and how to stop its spread. As newer variants of the virus are discovered, it has become even more important to enforce the various steps required to curb its spread. We can only fight this virus by wearing masks, using sanitizers, and social distancing. This paper proposes a hybrid masked face detection model for implementing the proper use of face masks. Our study focuses on combining machine learning models and Neural Networks. Even though various models have been proposed in the past for face mask detection, we tried to change the conventional machine learning methods by creating hybrid models like ResNet50 and VGG16 and combining classical machine learning models like SVM and Gradient Booster, and Neural Networks and comparing their performance. The Hybrid model architecture consisting of ResNet50 + SVM significantly outperformed the other models, returning an accuracy and precision of more than 97 and close to 100% each respectively.