On the savannas of Mpala, Kenya, a drone glides silently above a herd of zebras. Within a fraction of a second, an algorithm back at the research station identifies one particular animal—female, approximately four years old, first documented here eighteen months ago. No collar, no tag, no physical contact required. Just stripes. RAPID, a novel AI tool developed by an international team of researchers, can now recognize individual wild animals the way humans recognize faces: instantly, reliably, and by the patterns they were born with.
The system—its name is short for Real-Time Animal Pattern Re-Identification on Edge Devices—promises to transform how conservationists track wildlife. For decades, biologists have relied on camera traps, drones, and sheer patience to monitor animals across vast and remote landscapes. The challenge is that zebras, jaguars, and giraffes don't stay put. They migrate, they disperse, they vanish into terrain that makes sustained observation nearly impossible. Knowing whether you're looking at the same individual or a new one has always required either invasive tagging or painstaking manual comparison of photographs—work that could take weeks for a single population.
RAPID changes that calculus entirely. Developed by the Flight Robotics Group at the University of Stuttgart's Institute of Flight Mechanics and Controls, in collaboration with Eötvös Loránd University in Budapest and the Max Planck Institute for Intelligent Systems, the algorithm treats each animal's coat like a fingerprint. The spots on a giraffe, the stripes on a zebra, the rosettes on a jaguar: all unique, all permanent, all readable by machine.
"We need a reference database first," explained András Zábó, a researcher at Eötvös Loránd University and first author of the paper published in Methods in Ecology and Evolution. The system builds this database over time, adding newly identified individuals as they're encountered. When a drone or camera trap captures new footage, RAPID extracts what Zábó calls "descriptor vectors"—mathematical profiles of each animal's pattern—and matches them against known individuals in seconds.
The results speak for themselves. Tested on four public datasets containing images of animals including Amur tigers, RAPID achieved accuracies between 89 and 99 percent. On footage the team collected themselves—drone videos of zebras at Mpala and camera trap videos of jaguars in Ecuador's rainforests provided by the Jocotoco Foundation—the system reached 93 percent accuracy for jaguars and 80 percent for zebras. Processing speed proved equally impressive: 40 to 60 images per second on a standard desktop computer, and roughly 10 per second even on a basic edge device with no GPU.
That last detail may matter most of all. Conservation projects often operate in remote regions with limited connectivity and power. RAPID runs on hardware that fits in a small drone or camera trap housing, without requiring cloud computing or specialized chips. It is open-source and modular by design, meaning park rangers or research groups can adapt it to their own equipment with relative ease.
"Our recognition system works even on hardware with very limited processing power," said Aamir Ahmad, who leads the Flight Robotics Group. For wildlife monitoring, that accessibility could be the difference between a tool that exists in a journal and one that actually gets deployed where it matters—in the field, under the sun, watching the animals no one else can see.