Skip to Content

Machine-Vision Drones Monitor Animals in the African Savanna

Managing wild animals in remote areas requires accurate estimates of their numbers. Machine-vision drones can help.

The Kalahari is a semi-arid sandy savanna that stretches across huge areas of Botswana, South Africa, and Namibia. It is home to a wide variety of large mammals including giraffes, ostriches, gnus, and various species of gazelle.

Food resources constantly change in the savanna as rainfall changes, from grazing pressure and as bush fires spread across the land. To avoid overgrazing, land managers must ensure that the number of grazers is matched to the availability of food.

That requires significant monitoring. The most common ways of estimating populations of large mammals is to count them from a helicopter or to set up camera traps that record their movements through specific locations.

But these methods have significant drawbacks. Camera traps can only record populations in a single place and helicopter studies are expensive and time-consuming.

Another option is to photograph the area using a drone. This produces large numbers of images covering vast areas of land. But there is a problem. Analyzing these images is hard. It requires trained human operators to devote large amounts of time to the task.

So land managers would dearly love to have a better way to analyze these images.

Enter Nicolas Rey at the Ecole Polytechnique Federale de Lausanne (EPFL) in Switzerland and a few pals who have trained a machine-vision algorithm to do the job instead. They say the algorithm vastly reduces the time required from expert humans and could lead to significant improvements in population estimates of large animals.

Their method is straightforward. They begin with a 2014 drone-mapping study carried out at the Kuzikus wildlife reserve on the edge of the Kalahari in Namibia. This involved five drone flights over the reserve with a camera that took 6,500 pictures of the ground. Each picture was 3,000 x 4,000 pixels at a resolution of a few centimeters per pixel.

These images show many large mammals but they are sparsely distributed. And that makes them time-consuming for humans to find.

Rey and co’s idea is that a machine-vision system can be trained to do the job instead. But training requires ground-truth results that the machine can learn from.

So an important part of Rey and co’s method is creating this ground truth data set using a crowdsourcing campaign. They asked 232 volunteers to study the pictures and draw a polygon around each animal they come across. Each image was seen by at least three volunteers and a maximum of 10. The average number of viewers was five. If more than half the viewers agreed, the team assumed they had identified an animal.

In this way, the volunteers found some 976 large mammals in 650 images. Human experts then reviewed the results, removing 21 spurious cases in a process that took just 30 minutes. The team then used these examples to train and test their machine-vision algorithm.

The results are interesting. The team found the algorithm performed best early in the day when the animals cast long shadows. “We conclude that flying in the morning and always at the same hour of the day can lead to better results,” they say. For the same reason, it was also better at spotting animals standing up rather than lying down.

Nevertheless, the system worked well. “The system achieves a high recall rate and a human operator can then eliminate false detections with limited effort,” say the team. So a human operator is still needed but with a vastly reduced workload.

That has implications for animal conservation in Africa and other large areas. “It shows that the detection of large mammals in semi-arid Savanna can be approached by processing data provided by standard RGB cameras mounted on affordable fixed wings UAVs,” say Rey and co.

That’s interesting work showing how relatively cheap drone technology and increasingly powerful machine-vision techniques can be applied in remote locations. Animal conservation in these regions should be easier and more effective as a result.

Ref: arxiv.org/abs/1709.01722: Detecting Animals in African Savanna With UAVs and the Crowds

Keep Reading

Most Popular

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

OpenAI teases an amazing new generative video model called Sora

The firm is sharing Sora with a small group of safety testers but the rest of us will have to wait to learn more.

Google’s Gemini is now in everything. Here’s how you can try it out.

Gmail, Docs, and more will now come with Gemini baked in. But Europeans will have to wait before they can download the app.

This baby with a head camera helped teach an AI how kids learn language

A neural network trained on the experiences of a single young child managed to learn one of the core components of language: how to match words to the objects they represent.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.