In recent years, biometric systems have been used to provide access control security, to help identify and recognize people. However, today due to the advance of the pandemic some of the biometric systems are considered as sources of transmission and contagion of COVID-19. This situation motivates us to the development of a facial recognition access control system through the Application of Convolutional Neural Networks (CNN) that, by not having physical contact with the devices, safeguards against COVID-19. A pre-trained VGG16 model was used and the new CNN model was trained with the Transfer Learning application. The recognition system was tested using a public database such as celebA, obtaining an accuracy of 84%, this isn't the best possible accuracy as some authors report better results. However, our objective was not to provide the best accuracy with random data, we only need to achieve good accuracy with the company's controlled data, our CNN model achieves an accuracy of 98% in controlled conditions with an average identification time of 80 milliseconds. It has a low implementation cost that allows it to be competitive in low-income countries, like Ecuador, compared to international costs of state-of-the-art systems.