Linz researchers are working on a drone system that, with the help of artificial intelligence (AI), can combine thermal images in such a way that missing people can be located under heavy vegetation. Last year, the team presented a drone that can also do this fully automatically. Now they presented in the trade journal Remote Sensing a new approach that can be used to detect fast-moving people or wild animals under dense canopy in real time.
Problems in densely forested areas
When looking for missing or injured people, thermal imaging cameras are often used by rescue organizations on helicopters. In densely forested areas, however, this quickly reaches its limits, because there the images of the difference between body heat and ambient temperature can only be generated to a limited extent. The problem is that the vegetation covers the ground too much or the sunlit trees have a similar temperature as the missing person.
The team led by Oliver Bimber from the Institute for Computer Graphics at the University of Linz relies on an image processing technique for calculating occlusions (Airborne Optical Sectioning, AOS), which is combined with artificial intelligence. Several individual thermal images taken by a drone prototype are combined into an integral image. This is then classified using deep learning methods – the AI evaluates whether the detected object is actually a person.
Software is already in use
In the meantime, the underlying software is already being used by the project partner ÖAMTC for search operations. The solution is freely available for blue light organizations and can be built into their drone systems.
Most of the time, injured people remain in one place anyway, but if the person you are looking for is moving, the search becomes much more difficult. The team is now countering this with a newly equipped drone that carries a ten-meter-long carbon boom. There are ten cameras at a distance of one meter each. These then take their pictures at the same time, which cleverly combined results in the effect of a flying lens with a diameter of ten meters, according to the researcher.
