Drones Count Wildlife in Africa

Fast and precise: researchers are using drones to count animals in a nature reserve in Namibia. At the heart of the process is image recognition software that analyzes aerial footage captured by the drones.

The challenge is considerable. Some African national parks cover areas half the size of Switzerland, as stated in a press release from the Swiss National Science Foundation (SNSF). Counting antelopes, rhinoceroses, and wildebeest is therefore a laborious task that can only rarely be undertaken.

With drones, however, large areas can be surveyed at low cost. More than 150 images per square kilometre are captured, which can then serve as the basis for counting.

Artificial Intelligence Evaluates Image Data

"By automating part of the counting process, we aim to make it easier to collect more accurate and up-to-date data," Devis Tuia is quoted as saying in the press release. Tuia launched the project, known as Savmap, in 2014 at ETH Lausanne. Today, Tuia — a former assistant professor at the University of Zurich — holds a professorship at Wageningen University in the Netherlands.

To evaluate the photographic material, Tuia and his colleagues rely on artificial intelligence (AI) based on machine learning. This means the AI must be trained accordingly.

In this case, the task involves distinguishing animals from shrubs and rocks. Using an algorithm developed by doctoral student Benjamin Kellenberger, most images containing no animals can be immediately eliminated. For the remaining photos, the programme flags objects that could potentially be animals, as the researchers report in the journal «Remote Sensing of Environment».

«This first phase of cleaning up the image material is the longest and most tedious,» Tuia notes in the statement. However, according to the researcher, this work can only be taken over by AI if no animal is to be overlooked. The software must therefore have a fairly high tolerance, even if this results in more false positives.

Drone counting is cheaper and more accurate

Once the AI has processed the drone aerial footage, a human takes over the final review. This semi-automatic method was developed in collaboration with biologists from the Kuzikus reserve in Namibia. The researchers there have been using drones from the Swiss company Sensefly since 2014 to fly over the reserve.

At first he had been sceptical, says Friedrich Reinhard, director of the reserve, when quoted. "The drones take so many images that it hardly seemed usable to me."

But thanks to the AI's processing, a single person can complete a full count of the wildlife in the roughly 100-square-kilometre reserve in about a week. This is cheaper than counting by helicopter and also more accurate. More on environment and nature conservation.