B.C. ecologist uses facial recognition software to track grizzly bears

Melanie Clapham has spent the last three years snapping images of grizzly bears at Knight Inlet, on the B.C. coast, using small camera traps housed in metal and strapped securely to the forest branches.

Three years and thousands of images later, the behavioural ecologist and postdoctoral student at the University of Victoria has partnered with two software developers living in Silicon Valley and a grizzly research centre in Alaska to develop facial recognition technology used to identify the bears.

"They don't have distinctive markings on their bodies," said Clapham, whose interest in this technology stemmed from the need to "identify and recognize individual bears over time" as part of her behavioural research over the last 11 years. 

Now, she says, the open-source Bear ID software can be used and adapted by anyone and could have huge implications for understanding the animals' behaviour and mitigating bear-human encounters. 

Technology based on human facial recognition

Ed Miller and his partner Mary Nyugen are the software developers from California who connected with Clapham in an online forum for conservation technology in late 2017. 

The pair were looking for photos of bears "for fun" as a way to learn more about recognition software, and so they connected with Clapham to offer their expertise in adapting artificial intelligence.

"The technology we're using is based on the same software [used] to recognize humans," said Miller, who added that human identification is far easier, as there are literally millions of images the software can learn from.

"We need (lots of) images of individual animals to tell the system which bear is which," said Clapham, who explained "deep learning" as the process where the software trains itself to recognize certain bears more accurately the more pictures it has. 

This is especially important, given that a bear's appearance can change dramatically throughout the year as its fur moults and its weight fluctuates. 

Claphams says BearID currently has an 84 per cent accuracy rate.

Many practical applications

Clapham said she hopes the technology will be adapted by municipalities, governments, non-profits — as many groups as possible — as it will allow people to understand animal behaviour, like how they move in and out of densely populated areas. It could also help researchers understand the movements of endangered species.

It can track bears as they move "in a similar way that a human is tracked through airports," she explained. From there, authorities could make better-informed land management and conservation decisions. 

It could also help mitigate conflict encounters between bears and humans.

"If you have a bear digging through garbage cans, and you set cameras up … is this just one bear or is this five different bears coming into the area?" Clapham said.

Dallas Smith, president of the Nanwakolas Council, a group of five First Nations from Vancouver Island and the B.C. Coast formed to make land management decisions, said he's very excited for First Nations to use BearID, after connecting with Clapham.

"The grizzly bear is an icon in our cultural heritage. It's always been important to work in harmony with them," he explained. "It's really helping us gain a foothold in taking over the management of grizzly bear interactions in our territories."

He said the "collective territory" is working to gather more images for the system.