Detection of facial masks using Artificial Vision and Neural Networks to avoid contagion of Covid-19

Abstract

The objective of this article was to seek the detection of the use of the mask in people through an artificial vision program, applying the neural network model, where it is identified if the person wears the mask, thus seeking a contribution to mitigation and reduction of cases of contagion of Covid-19. In addition, the Python language is used together with Frameworks such as TensorFlow for the execution of trained neural networks and libraries such as Keras and Sklearn, used mainly in the learning process. Part of the methodology focused on the download and cloning of materials, creation and training of the neural network, the preparation of Anaconda together with Jupyter Notebook for the verification of the system. In the results we found that the programming correctly detected through a local medium, using Google Colab, for which a pre-established image bank was taken as a reference and the recognition was carried out. On the other hand, Jupyter Notebook is used for real-time video detection. Finally, it is concluded that it was possible to detect two types of images of people who are wearing or not wearing a mask with the variables "mask" and "no_mask", by training neural networks with a batch size of 8, steps of 50 and epochs of 25, with a classification loss result of 0.2120.