Finding people lost in a dense forest is a bit like finding a needle in a haystack. Something that can make it easier is to use heat-sensitive cameras that record body temperature. Sending up drones with thermal imaging cameras is a quick and inexpensive way to search hard-to-reach areas , but today only in 25 percent of cases do the cameras provide images of sufficient quality to be able to determine that there is a human in the image.
The problem is especially big on hot days when there is not much difference between the body temperature and the ambient temperature. It will be especially difficult in densely populated areas where the cameras find it difficult to penetrate the vegetation. It is also in these inaccessible areas that the need to be able to search from the air is greatest.
Now a research team from Johannes Kepler University in Austria has developed something that can provide a solution to the problem. They have used deep learning to teach an AI to detect humans in thermal images . The result is an AI that finds the right one in up to 95 percent of cases.
Simply put, the method involves the AI examining many images from the same area. The AI can then merge the images so that it sees it as a single large and detailed image. Further analysis gives the image a greater depth so that the treetops become blurred while what is on the ground is much clearer.
In the future, automatic drones will be able to search large areas themselves. Instead of having to send the images to a human on the ground who is trying to analyze them with difficulty, the AI can directly show where there are humans.
