FeatureModelling & Visualisation

Automatically detect and count animals on aerial imagery

Press Release 11 Feb 2020 - 17:28 SAST

Automatically detect and count animals on aerial imagery

Press Release 11 Feb 2020 - 17:28 SAST

Information from Picterra

Measuring animal populations is often the first step in conservation efforts for taking well-prepared, effective action. The problem researchers face is how to detect and count animals in an efficient and environmentally-friendly way.

Traditionally, wildlife surveys have been conducted directly by researchers, often using vehicles. Despite being performed in good faith, this way is always invasive and stressful for animals. Drones are considered game-changers and a growing number of researchers are flying them over specific areas, taking high-quality photos and then using them to detect and count animals. The problem is that when done manually, this counting can take an enormous amount of time and resources.

That’s why researchers from the University of Santa Cruz, California, have used Picterra to let machine learning detect and count animals from orthophotos taken by drones. The researchers have flown drones over Año Nuevo Island, which is a part of the Año Nuevo Reserve, off the shore of California. Their goal was to analyse the population of seals and sea lions, collectively known as pinnipeds because climate change has a significant impact on their habitats and behavioural changes.

Fifty flights delivered fifty orthomosaics covering the whole island at different dates. To start, researchers analysed 25 photos. Manually, it would have taken them 17,5 days to detect and count animals. Using Picterra’s machine learning tools, they’ve obtained accurate results in five hours. Analysing all 50 images took them approximately 8 hours, instead of 35 days of work.

Roger Fong, Picterra’s computer vision engineer, described the workflow in detail in this step-by-step tutorial: “How to prevent animal extinctions with drones and machine learning”. The final shared project results can also be found here.

In short, Picterra counted ~140 000 seals and sea lions across all 25 images and researchers could clearly see the changes of the pinnipeds’ population in 2018. Similar results can also be achieved with cows, sheep and different farm animals.

Using automated tools such as Picterra’s machine learning can help make detection operation efficient, even without coding knowledge, and can be replicated for other objects too.