Body Part Detection in Smoky Environments with Thermal Camera Using Deep Learning

Abstract

Human victim detection in smoky indoor environments during search and rescue missions is still challenging. This situation is due to the fact that fire fighters are on the one hand exposed to unstable building structures and on the other hand their cognitive fatigue, due to long search missions, reduce the efficient victim detection in these hazardous environments. In this paper, an approach to detect victims in real time with a thermal camera assisting fire fighters in their search and rescue mission, is presented. Thereby, a low resolution thermal camera is mounted on a remote-controlled mobile robot with a human hand detection using deep learning and display the detection in real time to an operator outside the danger zone. Experiments show that this approach enables an efficient victim detection in smoky indoor environments. The human hand detection model achieves a real time detection rate of above 90% in a dense smoke indoor environment.