O emprego de redes neurais convolucionais na identificação de máscaras de proteção no contexto da COVID-19

Abstract

During the Coronavirus (COVID-19) pandemic, the use of protective masks proved to be an im- portant resource to reduce contamination between people. The mask is a non-pharmacological, low-cost resource, but with great impact. Even so, some people do not use it, or do it incorrectly, so identification and guidance in these cases is important. This work presents a proposal for the use of Convolutional Neural Networks (CNN) as a way to identify people who are using a protective mask correctly, incorrectly or are without a mask (3 labels). For training and testing, a dataset with artificially generated images was used. Two CNN architectures were proposed, generating 5 models each. These models, despite having a simple architecture, achieved an accuracy greater than 99% in the test set. The architectures presented can be used for access control in indoor environments, as a way of guaranteeing the correct use of the protection mask.