Face mask detection using convolutional neural networks and haar cascade classifiers

Abstract

This project aims to develop a system that relies on face mask detection. It can be used as a method to control access to buildings, offices or any closed facility or public gathering places that promote human interactions. The access control is achieved through the monitoring of the people entering a certain building through a camera and decide whether to grant access or not to the person wishing to enter. The decision is based on whether the person is wearing a mask or not. The implementation of this system is made possible using two different machine learning techniques, namely: Convolutional Neural Networks and Haar cascade classifiers. The Haar cascade classifier is used to detect faces off frames captured from a video stream. The faces captured by the classifier are then fed to the CNN to classify whether the person is wearing a mask or not. The CNN architectures used in this project are the MobileNetV2, EfficientNet-B0 and a small custom CNN. These models are evaluated and compared to each other using various metrics in order to pick the one that suits well this specific project.