We can see from the data that koalas are in trouble here. They've recently been listed as endangered in Australia and Southeast Queensland was formerly a hot spot. The driver of that decline has been urbanisation. There is a strong link between mortality caused by cars and the breeding dynamics and the population dynamics of koalas. 

During the breeding season, we get a giant uptick in the number of koalas coming into care. That's why we deployed an AI powered camera network. The cameras will be automatically triggered by koala movement. Then the captured videos and images are automatically sent to a server at Griffith University. 

Then we develop an AI program to automatically analyse these videos and images to recognise the callouts. I think every day from these 24 cameras that we have deployed we can receive several hundred videos.

It's very difficult for us to watch each video to figure out what's there and especially a lot of the videos are force triggered by the movement of the trees or branch or leaves or rains. So we need some automatic analysis capability to detect whether there are animals.

We're using a deep neural network to detect the animals and this is a very up-to-date cutting-edge technology and this is very cool right now in the videos and images that we capture is much more advanced the technology that we use now to detect the animals.

It works like the human brain, right?

When we want to recognise something, we need to see the object repeatedly in order to train our brain to do the recognition. The problem for the koala recognition is we don't have enough images or samples to train the neuronet network. So therefore, we adopted a learning method called transfer learning. We use those neuronet networks train to detect human faces and then feed the neuronet networks with koala face images. 

So in this way we converted the human face recognition system to a koala face recognition system. So far we have achieved the recognition rate for about 90 to 95%.

It's a joint effort from a lot of partners include the Redland city council and Queensland rail. 

They helped us to provide the locations where we should put the cameras to check the animals and also the Daisy hill koala centre and dream world. We can put our cameras in their zoos or sanctuaries to capture the koala images and videos. 

So these images have been used to train our artificial intelligence algorithms.

I think what's been surprising is just the volume of images that's required to be able to allow the artificial intelligence to identify the koalas. A research project like this helps us to understand how that technology can be applied not only here in the Redlands, but where appropriate in other locations given that Queensland Rail operates a really large network.

We're all here for one goal, and that is conservation. We're all about sharing information and it's really good to be able to put people in contact, be that conduit between researchers so that everyone can get to the same goal faster and because it's technology based, it doesn't rely on human involvement. So, it's set and forget and away you go. 

You know, the technology does all the work for us and I think that's really important moving forwards because obviously to have people out there doing that job is just not possible. It's a much better experience in a sense that we're getting a lot of captures of different animals, not just human beings but koalas, Wallabies, foxes, Echidnas and so on.

This technology can then be used to identify individual threats. Could just be certain individual predators that is causing a problem. But to be able to identify that using this technology is a great step forward.

There may be issues with koalas crossing road lines, crossing rails. Perhaps with this increased knowledge about how koalas are crossing roads, we can better inform mitigation and management so that we can ensure a better long-term future for koalas.