Application of Deep-Learning Methods to Real Time Face Mask Detection
Diego González Dondo, Javier A. Redolfi, Gastón Araguás, Daiana García
Abstract
Due to the high rate of infection and the lack of a specific vaccine or medication for the new disease known as SARS-CoV2, the World Health Organization (WHO) has recommended the use of Personal Protective Equipment (PPE) as the main measure to avoid or reduce infections. One way to maximize compliance with this recommendation is through an automatic system that can recognize in real time whether a person is correctly using the corresponding PPE. This work presents the design, implementation and performance analysis of a system for recognizing the use of masks from image sequences, with the ability to operate in real time. Based on a generic object detection network, a training scheme is proposed for a detector of faces with masks and faces without masks, wherewith an average detection accuracy higher than 90% is obtained. This accuracy can be improved by using a network with a greater number of parameters, but with a longer computation time. The performance of the detector is validated with video sequences of people with and without facemasks, captured in different environments.