Attaching cameras to animals has allowed wildlife film-makers and biologists to capture some extraordinary footage, from bird’s eye views of eagles soaring in the skies to sperm whales hunting in the depths. Now artificial intelligence could help us capture a lot more revealing footage.
A team in Japan has created a low-power AI system for recognising when animals are doing something interesting and switching on high-power systems like videos only during those moments. That means the devices can keep working for far longer before running out of power.
Takuya Maekawa of Osaka University worked with biologists using biologgers — devices equipped with a camera, accelerometer and GPS— to study black-tailed gulls from a breeding colony located on Kabushima Island. The devices weigh just 27 grams and cannot record video for long.
Advertisement
Maekawa trained an AI to recognise when the birds were feeding by analysing the movements recorded by the low-power accelerometer. The AI checks whether the bird is flying and whether the patterns of movement resemble those associated with feeding, based on previous recordings.
The team then put the system to the test. Of 27 videos taken without the AI, none showed feeding. In fact, the birds were stationary in 24 of them. By contrast, of 185 taken with the AI, the birds appearing to be foraging in 58 of the videos, and were stationary in just 41.
That’s a success rate of around 30 per cent, which is impressive for an algorithm that can run on the tiny processing unit inside the biologgers. This first small study even recorded some previously unknown behaviour, with several videos showing the black-tailed gulls catching insects out at sea.
“Feeding on insects over the sea has not been observed before,” says Maekawa.
bioXriv
