AI is being used to identify animals, plot their movements and spot wounds in a bid to help conservationists by cutting down the time taken to sift through video footage from camera traps.
at Liverpool John Moores University in the UK and his colleagues first began developing their AI model three years ago and have successfully identified 44 different species in motion-activated camera footage, as well as in overhead images from drones and audio. They are currently trialling the latest version of the system at Knowsley Safari, a 222-hectare safari park in England, to test its ability to identify giraffes, rhinos and lions before wider-scale tests in the wild in South Africa, which were delayed by the covid-19 pandemic.
The researchers have previously put their algorithm to work for the UK’s National Health Service, identifying bed sores in older patients from digital images. They applied this same algorithm to animals, using it to identify whether they had any wounds from video and drone images. “A wound is a wound really,” says Fergus. “It works just as well on the animals as on the patients.”
Advertisement
To keep accuracy high and the models simple, the team offers conservationists a different software model for each continent, each with accuracy of up to 95 per cent, according to the researchers. The models are trained on 66,000 images of 44 species, including European hedgehogs, foxes, red and grey squirrels, and birds such as the blue tit and common wood pigeon – as well as humans. The group offers the software and servers to power it for free, and has more than 100 active users around the world, including in South Africa and South America.
With 4G-enabled cameras and drones, the footage can be streamed anywhere in the world, then analysed and presented to conservationists via a web portal within 4 seconds. This can also provide real-time ability to stop poachers in their tracks, rather than simply find evidence of them months or even years later after footage has been manually analysed.
at Knowsley Safari says that on one conservation project she participated in outside the park it took her months to manually classify animals in 7000 hours of video footage, but when the same data was sent to the AI researchers as a test to check results it was processed in a day.
“It’s fabulous and it’s very exciting, but it’s also incredibly frustrating if I think about how many hours I sat at the computer,” says Walsh. “It means we’re freed up to do more exciting things as opposed to sitting staring at a computer. We can do more and achieve more.”
Sign up for Wild Wild Life, a free monthly newsletter celebrating the diversity and science of animals, plants and Earth’s other weird and wonderful inhabitants
Article amended 27 January 2022
We corrected the number of images used to train the models
