Tufts engineers and conservationists are helping track and protect wild elephants in East Africa
If you want to track an elephant in the wild, it’s not easy. Traditionally, you have to attach a radio collar to it, which might interfere with the elephant’s natural movements.
Now School of Engineering Professor Karen Panetta is developing drones equipped with a novel artificial intelligence system to track elephants in their natural habitat—and is already deploying them in the field in East Africa. As she and her students continue to develop AI for the project, she hopes the technology will allow conservationists to collect high-quality data that they can use to make wildlife management decisions.
Panetta says her Visualization, Sensing, and Simulation Lab creates “artificial intelligence engines” that are then installed into various machines, including drones. Her lab has been working for years on using drones to aid in disaster recovery, assessing damage from the sky.
She was approached about four years ago by members of the Tufts Elephant Conservation Alliance, who asked if the drones could be used to aid wildlife conservation. Intrigued by the opportunity to expand the uses of her research, Panetta jumped at the idea.
Panetta and her team developed an AI-equipped drone that recognizes individual elephants from images, even when those images are dark, seen from afar, or otherwise difficult for conservationists to use effectively. Knowing where specific elephants are can help wildlife conservationists determine ideal habitats and keep animals safe from interactions with humans, which can be deadly.
Since elephants frequently travel at night, the AI is trained to identify elephants with thermal imaging—a method that detects elephants’ body heat. Obafemi Jinadu, a second-year doctoral student in Panetta’s lab, says the team has had “huge success” identifying elephants in crowded images and in the dark.
The lab’s technology can also collect data on individual elephant health, by identifying skin infections, changes in elephants’ size, or altered behavior. For example, when an elephant is upset or stressed, it may raise its ears, which could be a helpful data point for conservationists monitoring the animals, according to Jinadu. “We try to get as much information as we can from the animal’s body language,” he says.
The drones might also deter poaching, Panetta says, because poachers will recognize drones as signs of human presence, and hopefully avoid the area.
Panetta says drone tracking is more humane for the elephants than other strategies, like attaching radio collars, which can interfere with the elephant’s natural behaviors. “The current approach for elephants now is quite invasive,” Jinadu says. “We’re trying to track migration patterns with a very non-invasive approach.”
“Our approach from the beginning was saying that whatever we do, we cannot alter the elephants physically,” Panetta says. “That was really important.”
Part of that non-invasive approach includes identifying elephants on a first-name basis from afar, an aspect of the AI that Jinadu is working to improve. All elephants, he says, look quite similar, especially since they’re free from markings, like stripes or spots, which help conservationists identify individuals from other mammal species. Right now, only people who have worked directly with specific elephants know them well enough to differentiate them.
Jinadu is hoping to change that by identifying the characteristics—like tears in ears, head size, and tusk shape and orientation, among others—that differ between individual elephants, they can use that information to train the AI to match those characteristics to elephant names.
“It’s a tricky task,” he says, and current monitoring technology has only about a 56% success rate. “But we’re positive we can do way better than that.”
Panetta’s drones are already used by conservationists at the Masai Mara National Reserve, a protected elephant habitat in southwest Kenya, where tens of thousands of elephants roam.
The team hopes that the technology they’ve developed could start to be used for conservation of other species, as well. For example, Panetta says that the AI could be particularly suited to track rhinoceroses. Additionally, the team has been collaborating with Allen Rutberg, a research associate professor at Cummings School of Veterinary Medicine, to determine how to adapt the technology for white-tailed deer conservation, since manually tagging white-tailed deer is labor-intensive and can be dangerous.
To Panetta, the possible applications of AI are limitless for helping humans and animals prepare for the unexpected—like a band of poachers or a change in elephant migration patterns. “There are so many different scenarios that we as humans can’t even begin to create experiments for,” she says. “AI allows us to do that.”