Koala populations could benefit if a pilot program in South East Queensland successfully uses “facial recognition” technology to track individual animals.
A team of Griffith University artificial intelligence (AI) researchers hopes the technology will help protect the declining population of native marsupials.
The goal of the project is to monitor the koalas’ road crossing behaviours and analyse how many animals are using above-road crossings or underground paths.
Researchers believe the information will inform council and state government development decisions and support the long-term conservation of koalas across the region.
Associate Professor Jun Zhou, from Griffith’s School of Information and Communications Technology, said the cameras would be at well-known koala road crossings.
Who’s that koala crossing the road?
Until now, researchers have had to capture images of koalas then examine each photograph to identify the koala.
“Now artificial intelligence has evolved to help humans do many tasks, including automatic animal detection,” said Professor Zhou.
“The idea is to develop a network of cameras where koalas cross the roads … send the images back and identify and automatically recognise the koala.”
Professor Zhou said similar technology was used by law enforcement and airport security to identify individuals.
He said there were three types of recognition tasks.
“First, we have to recognise whether there is an animal in the image or not,” he said.
“Secondly, we need to detect whether this animal is a koala or some other animal.
“The third one, which is more challenging, is to distinguish individual koalas.”
Telltale koala bottoms
Professor Zhou said it was difficult for untrained observers to identify individual koalas but experts relied on features such as “how big or floppy the ear is” or if there was “a particular pattern on the nose or some specific pattern on its bottom”.
“With our recognition system we will definitely use this human knowledge to help the development of the AI system,” Professor Zhou said.
Population ecologist Dr Douglas Kerlin said the project was funded by the Queensland government and researchers were working with Redland City Council, which has helped identify koala habitats and road crossing points.
“We initially want lots of images of koalas,” Dr Kerlin said.
Researchers have also been speaking with Brisbane City Council, which has a number of wildlife crossing road infrastructure projects.
That includes a land bridge linking Karawatha Forest on the southern side of Compton Road to an area north of Compton Road known as Kuraby Bushlands.
Dr Kerlin said the technology could also be potentially used to monitor disease in a koala population
“Road-crossing behaviour is only one of a number of applications for this kind of a system,” he said.
“You could use it to monitor four koalas at a location, for instance, prior to property development.”
Easy to miss in the wild
Dr Kerlin said current methods of identifying koalas were labour intensive.
They involve either spotting the herbivores in bushland or catching koalas and attaching radio collars, which can impact on an animal’s welfare.
“Koalas are quite cryptic and they live right up in the tops of trees so they are easy to miss,” he said.
“It’s a passive surveillance system that has a lot of positives down the track … to help animals in the whole region.”
Professor Zhou said the AI system needed many images of koalas so the technology would be deployed in koala crisis centres to help train the technology to distinguish koalas based on their appearance and movements.
Conservation groups involved include Koala Action Group, Currumbin Wildlife Sanctuary, Daisy Hill Koala Centre and the Moggill Koala Rehabilitation Centre.
“At the same time we plan to deploy some QR (quick response) codes in these koala centres so the visitors can scan the QR codes and upload their images to our server,” he said.
The pilot program involves 20 cameras being placed at koala crossing locations in the Redland City Council area by the end of July.