Improving Thermal Camera Performance in Fever Detection during COVID-19 Protocol with Random Forest Classification

Abstract

The AMG8833 sensor can be utilized for a low-cost thermal camera-based body temperature measurement during COVID-19 protocol enforcement. However, the sensor is not accurate enough for body temperature measurement, so fever detection performance becomes poor. The aim of this study is to apply Random Forest as a classifier in a thermal camera body temperature measurement that uses the AMG8833 sensor and evaluate its performance in detecting fever. In addition to the AMG8833, the thermal camera made also uses a webcam for face detection, and a Raspberry Pi as a minicomputer and a place to apply the Random Forest model. That way, the Thermal camera undergoes three processes, namely face detection from the image captured from the webcam, then temperature and fever detection from AMG8833. From the receiver operating curve (ROC) test conducted, Random Forest area under curve (AUC) value is superior compared to the Logistic Regression and Decision Tree methods with a value of 0.977. Furthermore, the sensitivity and specificity values of Random Forest in detecting fever are 88.5% and 99.5%, respectively. This value is higher than a detection system that does not use Random Forest classification for fever detection.