On the trail of wildlife with drones and AI

Even the most experienced wildlife biologists find it difficult to estimate how many animals roam the forests. The animals move quickly, are shy and masters of camouflage. Their counting is correspondingly complicated. David Schedl, professor of visual computing at the Hagenberg campus of the University of Applied Sciences in Upper Austria, wants to solve the problem using new technologies. For a year now, collaborative partners have been working on the research project “BAMBI” (an acronym for Biodiversity Airborne Monitoring based on Intelligent UAV Sampling).

Drones take photos, AI recognizes animals on them
Camera drones take aerial photos of areas that may be home to wildlife. At the same time, self-learning algorithms are trained to automatically recognize and track animal species. With the help of these selective flights, a statistically more accurate estimate of the population of the various animal species will be possible in the long term.

But how is the camera supposed to capture what’s happening on the forest floor when the treetops obscure the view? “That’s the new thing about our project: in addition to the already known thermal imaging cameras, we are using light field technology for wildlife monitoring in combination with drones for the first time. It is a first field test to find out how well this can actually work,” says Schedl. Because while a conventional camera only takes pictures, light field technology uses considerably more data. Instead of a single image, a light field consists of many images. A point on the ground is thus visible several times from different positions of the drone After processing all data, an average value is calculated from the images, the opacity can be resolved and wildlife can be made visible to the camera and classified.

For data processing, it is also important to know the precise position of the drone and the nature of the terrain over which it flies, with its heights, depths and irregularities. “We are building on elevation data obtained from laser-scan aerial surveys, which are available almost everywhere in Austria,” says Schedl.

AI detector for deer, deer, wild boar and chamois
A crucial part of the research project is to generate the necessary large amount of data of sufficient quality under the additional conditions for light field recordings. The artificial intelligence that is being developed must learn to correctly assign the different animal species using image processing. It starts with deer, deer, wild boar and chamois.

The project will run until April 2025. In order to obtain as much data as possible for training the AI, a number of naturally structured enclosures from animal parks were flown in as a first step, for example in the city of The Hague or in the Cumberland Zoo. Footage has also been made in the wild.

Animals must not be disturbed during the count
The camera drones usually fly at an altitude of 30 to 40 meters. “We want to disturb the animals as little as possible. Our goal is to count them,” says Schedl. The intention behind this is to prevent threatening ecological problems, such as the loss of biodiversity, through continuous monitoring. In the beginning, it’s usually about regulating the wild population, if needed. This would also allow rare animal species such as lynx or wolves to become the center of attention for the purpose of species protection.

“BAMBI” is funded by the Austrian Research Promotion Agency (FFG). The cooperation partners on board are the Bureau for Natural Ecology and Forestry, the geodata service provider Umweltdata and the drone specialist ViewCopter. Support comes from the hunting associations in Lower and Upper Austria and Tyrol.