Application of Deep Learning for Facial Recognition Obstructed by Face Masks and/or Glasses

Deep learning allows a machine to learn autonomously using neural networks. This study applies deep learning for facial recognition even when using masks and/or glasses. The research uses Python and the pre-trained convolutional network library VGG16. The application environment requires a dataset where the photographs of the participants are stored, which amount to a total of 1000 images, essentially capturing the breadth of the face using a mask and / or lenses; and placed in three subfolders: Train, Test and Valid. In the first phase, the training is carried out, learning in 500 epochs to create trained model. In a second phase, facial recognition is performed with the presence of occlusions on the face; using a webcam. The accuracy or precision achieved in the training of the neural network in Google Collab is 0.1 and the percentage of success obtained from the application in the second phase is 84.68%. The project manages to recognize people when they use a mask, glasses, mask and glasses or without the use of these; with the success rate mentioned above