Automatic Aerial Victim Detection on Low-Cost Thermal Camera Using Convolutional Neural Network

The first thing to do by the search-and-rescue (SAR) team after the disaster occurred is to find the location of the victim quickly. Thus the loss of lives can be reduced. After the disaster, the environment usually very messy, containing debris from building, soil, and gravel, which makes it harder to find the victims. By detecting the temperature using a thermal camera, it can easily be distinguished between the victims and the other background. Previous work, the technology to detect a person using a thermal camera from aerial has been developed, but it is only working with the most nearly uniform background. In this paper, we developed an automatic aerial (drones) victim detection using a thermal camera. A low-cost thermal camera has been used so that anyone can quickly implement in the real situation. By combining CNN as its algorithm that widely uses for its excellent performance on object detection, it can easily detect victims from the low-cost thermal camera and distinguished from complex background. Experiments show very well that the proposed method able to detect victims from aerial thermal view with accuracy AP = 82.49%. We believe it could bring benefits for future work with the related field and able to help search-and-rescue team to find and evacuate the victims quickly.