Facial recognition and body temperature measurements based on thermal images using a deep-learning algorithm

Abstract

Recognizing the early symptoms of the SARS-CoV-2 virus (COVID-19) is essential for minimizing its spread. One of the typical symptoms of a person infected with COVID-19 is increased body temperature beyond the normal range. Facial recognition can be used to separate healthy people from those with high body temperatures based on thermal images of the faces. In this study, the XEAST XE-27 thermal imager modes 2, 3, and 4 comprising 1500 thermal images each were compared. The facial recognition was performed using a convolutional neural network. Additionally, body temperatures were extracted from thermal images using matrix laboratory (MATLAB) by considering the minimum and maximum temperatures of each mode and class. The network training results indicate that the accuracies achieved by the proposed facial recognition system in modes 2, 3, and 4 are 87.33%, 92.33%, and 91.66%, respectively. Furthermore, the accuracies of body temperature extraction in modes 2, 3, and 4 are 70%, 60%, and 40%, respectively. Thus, the proposed system serves as a contactless technique for the early detection of COVID-19 symptoms by combining facial recognition and body temperature measurements.