Facial Mask Detection Using Stacked CNN Model

Abstract

The coronavirus outbreak has affected the whole world critically. Amongst all other things, wearing a mask nowadays is mandatory to avoid the spread of the virus according to the World Health Organization. All the people in the country prefer to live a salubrious life by wearing a mask in public gatherings to avoid contracting the deadly virus. Recognizing faces wearing a mask is often a tedious job as there are no substantial datasets available comprising of masked as well as unmasked images. In this paper, we propose a stacked Conv2D model that is highly efficient for the detection of facial masks. Such convolutional neural networks work effectively as they can deduce even minute pixels of the images. The proposed model is a stack of 2-D convolutional layers with relu activations as well as Max Pooling and we implemented this model by using Gradient Descent for training and binary cross-entropy as a loss function. We trained our model on an amalgam of two datasets that are RMFD (Real World Masked Face Dataset) and Kaggle Datasets. Overall, we achieved a validation/testing accuracy of 95% and a training accuracy of 97%. In addition to this, we also developed an email notification system that sends an email whenever a person is entering without a mask and it will also prompt the user to wear the mask before entering into the system. Such a system is beneficial to large multinational companies and can be deployed there as the spread of viruses there is high because employees are from different regions.