Get access to everything we publish when you sign up for Outside+.
Bearpacker is Backpacker’s annual celebration of bear safety, science, and stories.
Facial recognition technology is used to unlock your phone, find missing persons, detect shoplifters, and even was used by Taylor Swift’s security to ID stalkers (seriously). But now, it’s revolutionizing bear conservation.
In fact, the brown bears of Glendale Cove in British Columbia have cameras around them 24/7. These camera traps—small, unobtrusive devices—track individual bears to gather information on their behavior. That’s right: Surveillance cameras are watching their every move, just like they watch ours.
How Bear Facial Recognition Came to Be
Bear facial recognition was the result of an unlikely collaboration between conservation scientist Melanie Clapham and software developers Ed Miller and Mary Nguyen. The trio met in 2017 through an online community while they were working on separate projects with the goal of identifying individual bears from photos and camera traps. After a few discussions, a collaboration began and they formed the BearID Project.
“We developed facial recognition technology for grizzly/brown bears so we can identify individual animals for research and monitoring,” says Clapham, a postdoctoral research fellow at the University of Victoria in Canada. “[BearID] can help estimate population size, which dictates management and conservation decisions regarding species and populations.” It can also track wildlife behavior and migration, providing a deeper understanding of the management and conservation of the species.
Working in two phases, the BearID application begins by running a facial-detection tool to measure features, like the forehead, ears, eyes and tip of the nose. Next, the system sorts and labels the images to decipher one bear from another. Whereas traditional monitoring involves capturing and tagging, a stressful situation for these furry beasts, camera traps can do the work with less human involvement.
How Well Does Bear Facial Recognition Work?
BearID is good, but it’s not perfect: The system currently has 84% success rate for ID’ing bears.
“Our ultimate goal is to develop a robust AI system that can identify individual animals that are difficult to recognize,” says Clapham. Unlike zebras and giraffes, bears have no identifiable markings, making it a challenge for most humans to discern one bear from another. What’s more, an individual bear’s appearance can change drastically throughout the year. Entering spring with a lean body, these grizzlies can gain up to half their body weight by hibernation time. Those changes in appearance have restricted research and monitoring techniques in the past.
At present, the BearID Project is focused primarily on brown bears, specifically in Katmai National Park in Alaska and Glendale Cove in British Columbia.
“We are working with Indigenous Guardians of the Nanwakolas Council to use the software to track individual bears across territorial boundaries using remote trail cameras,” says Clapham. The footage gathered will support the process of retraining machine learning to apply its use in data processing and analysis.
Although scientists consider the brown bear population stable, the researchers hope they’ll be able to use the tech they’ve developed for the project to help protect more endangered creatures as well.
“We are interested in how this technology could be developed for other species, specifically threatened species such as caribou and other bear species like polar bears and sloth bears,” says Clapham. “We have also received interest from organizations looking to use BearID to better understand human-wildlife conflict and coexistence, including black bears and grey wolves.”
Got a Bear Pic? Join The Cause
If you’re reading this, you can help further the BearID team’s research by sending in your face-on bear photos (or videos) of identified wild or captive bears.