Artificial intelligence-based facial body temperature measurement system using thermal image and YOLOv4

ABSTRACT

Due to the recent spread and long-term continuation of COVID-19, numerous non-contact body temperature measuring equipment is being introduced in public places to prevent cross-infection. Therefore, extensive research on the non-contact body temperature measurement method, using a thermal imaging camera, is being conducted globally. The existing method of measuring the body temperature using a thermal imaging camera has several limitations including a severe temperature deviation while measuring the body temperature, blurring of the location of the set measuring point, or not obtaining an accurate measurement due to obstacles. To overcome these limitations, we used deep learning to detect faces in thermal images, and measure the body temperature using a multipoint-based image processing with histograms of the detected areas. The proposed deep-learning-based method exhibited several advantages such as higher accuracy, wide area coverage, and efficient detection irrespective of obstacles. Compared to the conventional body temperature measurement, that proposed in this paper resulted in an average body temperature measurement error rate of less than 2%. © 2021, Institute of Control, Robotics and Systems. All rights reserved.