Thermal Face Detection for High-Speed AI Thermometer

Abstract

In the era of COVID-19, temperature measurement becomes a crucial procedure for protecting public spaces against the virus. Artificial intelligence techniques such as object detection deep neural networks (DNNs) have been adopted to enhance the accuracy of contactless temperature measurement. However, the computation-demanding nature of DNNs, along with the time-consuming fusion of video and thermal camera frames, raises hurdles for the cost-effective deployment of such AI thermometer systems. In this work, we propose a high-speed and cost-effective implementation of an AI thermometer. We develop a thermal face detection network to detect faces for temperature measurement without a video camera. We optimize the proposed network's precision and structure to exploit high-throughput reduced-precision computations available in the embedded AI platforms. The resulting AI thermometer system demonstrates a live temperature measurement with a speed of 160 frames per second.