Human Object Detection using Very Low Resolution Thermal Cameras for Urban Search and Rescue

In this paper, we propose an innovative assistive system for victims' localization under an Urban Search and Rescue environment. The localization exploits miniaturized Long Wave Infrared (LWIR) cameras and algorithms properly combined with template matching schemes from visual cues. The approach first thresholds the data to roughly estimate alive victims' location and then apply a segmentation scheme to improve the probability analysis. The key challenge of our approach is the fact the thermal image used captures images at very low resolution. This selection is necessary so that the sensor is light and thus able to be carried out by a snake robot. Experiments are presented to demonstrate the efficiency of our results both at the laboratory and under real-life conditions. The latter have been conducted under the aegis of FP7-INACHUS European Union funded project