By Donna Lu
Arsha Nagrani at the University of Oxford and her colleagues have developed a facial recognition AI that can detect and identify the individual chimpanzees captured in video footage recorded in the wild. Using the AI, they can cut down the time and resources needed to track animals in their natural habitat.
The algorithm could help researchers and wildlife conservationists study the complex behaviours of chimpanzees and other primates more efficiently.
The team trained the AI on 50 hours of archival footage – spanning 14 years – of chimpanzees in Bossou in Guinea, West Africa. The footage of 23 chimpanzees, with estimated ages ranging from newborn to 57 years, yielded 10 million facial images.
The algorithm learned to continuously track and recognise individuals from raw video footage, says Nagrani.
It performed well even on low light and poor-quality images, and worked for images in which the chimps weren’t looking towards the camera. The AI had an overall identity recognition accuracy of 92 per cent, and correctly identified an animal’s sex 96 per cent of the time.
To compare its ability with that of humans, the team then selected 100 random still images and tasked the AI as well as people with identifying the chimpanzee in each image.
The algorithm achieved an accuracy of 84 per cent, taking 30 seconds to complete the task. In comparison, researchers who were experienced in recognising the chimps took 55 minutes and had an average accuracy of 42 per cent.
The algorithm will allow researchers to more efficiently examine how behaviour and social interactions change over years and generations of animals, says collaborator Daniel Schofield. “You can start to build up a social network,” he says.
By quantifying the interactions between individuals, they were able to track changes in community structure over time.
Though the team trained the AI on chimpanzees, it could be applied to other primates, says Nagrani.