People leave calling cards wherever they go in the form of unique fingerprints. As every Sherlock Holmes fan can attest, they provide a crucial means for law enforcement to determine someone’s identity. But it turns out that humans aren’t alone in having diagnostic prints. An Alberta wildlife reserve is turning to artificial intelligence to identify individual black bears based on their paw prints. The footprint identification technology (FIT) provides an inexpensive and non-invasive way of tracking wildlife, and the technology may assist researchers in determining what wildlife populations are at risk and need conservation help.
“When you’re doing wildlife rehab, pretty much all the animals you receive in a wildlife sanctuary have been injured or orphaned through human activity,” says Simon L’Allier, a biology intern at Cochrane Ecological Institute and a Master of Science candidate at York University. “When it’s human-caused, it’s our responsibility to get these animals back on track.”
One of the major bonuses for using FIT is that the process is simple and low cost. “All you need is a ruler and a camera. A smartphone camera can do the job quite well,” says L’Allier.
One hitch? Getting the project up and running requires the researchers to “train” the FIT software to recognize individual black bear pawprints. And, turns out, “you can’t train the model with just random tracks you find in the wild,” says L’Allier. When you capture photos of prints, you need to know the individual, the sex, and the age of the black bear who made those prints. This information is sent back to Wildtrack, who teaches the artificial intelligence model to recognize individuals in the database from a single paw print.
Once the model is trained, it will be able to distinguish pawprints from bears that aren’t in the database. “It can tell you: this is a bear track, and this is from an unknown individual because it’s not in our system,” says L’Allier. “You add it to the database, so once you see that track again in the wild for a second time, you know you’ve seen this individual before.”
The ability to recognize new bear paw prints and add them to a database provides a huge advantage over other monitoring technology. While GPS collars have their uses and can provide fine data readings like an animal’s heartbeat, “with GPS collars, you have to capture every single bear in the area and put a collar on them,” says L’Allier. In contrast, FIT doesn’t require researchers to tranquilize or handle the bears.
Typically to train the AI model to be accurate, you need data from twenty individual black bears, says L’Allier. The model requires twenty photos of each bear’s front left paw print. Gathering that many clear photos can prove tricky. “Bears walk in their own tracks,” says L’Allier. “The back foot will go on the front paw track and so forth.” This can distort the shape of the track.
The ground cover bears trek across affects the quality of the paw print. Coarse sand is an ideal substrate for making clear paw prints, says L’Allier. Mud doesn’t produce great tracks, because it sticks to the bear as the paw pulls back, he adds. You run into the same problem with snow.
There is an exception: the snow on a “late afternoon when there’s a lot of sun and it starts melting on the surface” can create readable tracks, says L’Allier.
L’Allier was able to begin the tracking project with a rehabilitated black bear named Siksi’naan who was released in early October. But with winter setting in and bears heading into hibernation, data collection is on hold until spring of 2021.
In the meantime, wildlife enthusiasts can contribute to WildTrack’s database by joining the Wildtrack Fast Data Collection project on the Epicollect 5 app and submitting their own photos. As a bonus, the app can identify tracks to species. Now you can finally settle if the mystery tracks around your cottage garbage can are from a raccoon or a black bear.