Performance Benchmark in Febrile Mass Screening Detection

Abstract

Recently, detecting and tracking people using infrared sensors in public surveillance has been given attention by many researchers since the global outbreak of severe acute respiratory syndrome, and machine vision is bound to play an important role. In machine vision, sliding window approach has appeared as most promising approach. Since Papageorgious et al who proposed the first sliding window detetectors until Dalal et al who have come out with a large gains detector based on histogram adoption, the ideas of sliding in human detections continue to serve as a root for modern detectors. While there is increasing research in thermal spectrum, very little research focus on the effectiveness of state of art detectors in febrile mass screening application. The exposed area of head-to-shoulder is an important body part for region of interest detection priorly before the temperature of febrile person is measured by the thermal camera. In addition, some of the thermal datasets used in previous research are more fit to surveillance or safety applications where the targets are mostly far-scales. One of the challenges in detecting the pedestrian in thermal images is the nature and quality of image in infrared spectrum as well as the real crowd situation in public area that cause occlusion. Therefore in this paper, we are interested to evaluate the top three detectors’ performance on thermal images taken during fever screening in Kuala Lumpur International Airport. We also analyze the best approach to be adopted in the detection using a new context of training and evaluation.