AI koala: facial recognition could help save the species

The koala is endangered in northeastern Australia – and one of the biggest problems facing the species is how to establish their numbers.

One way to tackle the problem is with local community groups, a camera system – and some AI.

“This is the current challenge faced by the research community: there’s no reliable technology that can accurately and continuously count the number of koalas in the wild,” says Professor Jun Zhou, from Griffith University, who is leading a pilot study on koala observation.

Koalas are hard to spot and frequently change their territory, making them fiendishly difficult to count manually, or tag and monitor – so it’s hard to tell which conservation efforts help them to thrive.

“The technology that we deploy, based on cameras and artificial intelligence technology, will allow a continuous monitoring capability in a certain area,” says Zhou.

Zhou and colleagues began the project in July 2021, setting out 24 cameras at “koala crossing locations” in the Brisbane and Redland city council areas.. The cameras were set up to automatically start recording when triggered by koala movement.

Then, a team of Griffith researchers and community koala monitoring groups used the images to train an AI, so that it could not only spot koalas, but recognise individual animals.

This has given them a tool that could be used to monitor koala populations in much bigger areas.

“For example, we could put 100 cameras in a one square kilometre research region, and it could monitor the whole area, 24 hours a day, seven days a week,” says Zhou.

“Our AI can recognize or analyse those videos to see whether they’re koalas and which koala it is. So in this way, we can count all the koala movements.”

This can help them spot new koalas in the area as well.

But the results aren’t all positive: early estimates from the Redlands project suggest that koala numbers are in trouble.

“Southeast Queensland was formally a hotspot. The driver of that decline has been urbanisation,” says co-researcher Dr Douglas Kerlin.

The team has just received another grant from the Queensland Government to expand the trial to 10 local government areas.

Zhou says the new trial will help improve the accuracy of the technology. He’s also hoping that more local councils will adopt it.

“But in order to get the technology adopted in a wider area, that will take time, definitely.”

Local council employees and volunteers are the best resource to deploy the technology, says Zhou.

“Although the data capture, data transfer, data analysis and reporting has been automated by our technology, we still need people to help deploy and maintain the camera network,” he says.

“That’s why we want to get the local city council and koala conservation communities involved so that they can help this process. Instead of us going to maintain this camera, we can get the community to help to deploy and maintain the cameras.”