Tracking down poachers in the pitch black of night used to require expensive military gear and exhausting human oversight. Then a high school student from New York changed the math. Anika Puri built a machine learning system named ElSa that spots poachers in nighttime drone footage with over ninety percent accuracy, and she did it using about three hundred dollars worth of off-the-shelf hardware.
If you spend any time looking at wildlife conservation technology, you notice a massive disconnect. High-end thermal infrared cameras cost upwards of four thousand dollars each. Protected reserves in Africa and Asia often operate on razor-thin budgets, meaning advanced surveillance remains out of reach where it is needed most. Puri looked at that financial barrier and decided to bypass it entirely.
The Flaw in Traditional Wildlife Surveillance
Most drone monitoring programs rely on high-altitude thermal imaging. At four hundred feet in the air, humans and animals look like tiny, indistinct thermal blobs. Traditional software tries to identify these blobs by analyzing individual video frames based on static shapes.
That approach fails miserably in the real world. Static shape detection in chaotic nighttime conditions can drop to twenty percent accuracy. Rangers sitting in front of monitors for hours miss things. Fatigue sets in. Poachers slip through the gaps.
Puri realized that looking at a single frame is the wrong way to solve the problem. Instead of asking what an object looks like right now, her software asks how that object moves over time.
How the ElSa System Works
ElSa, short for Elephant-Savior, evaluates spatio-temporal patterns in thermal infrared video sequences. It tracks subtle differences in speed, turning radius, directional behavior, and whether a moving target travels alone or alongside a herd.
By analyzing these movement signatures across multiple successive video frames, the algorithm separates human behavior from animal behavior with remarkable precision. When tested on the BIRDSAI thermal infrared dataset sourced from national parks across Africa, the system hit 90.9 percent accuracy. That represents roughly a fourfold improvement over traditional frame-by-frame detection methods.
Spotting a human in a protected wildlife reserve doesn't automatically mean a poacher is present, but it instantly flags suspicious nocturnal activity for rangers who can't watch every monitor at once.
Democratizing Conservation Through Low Cost Tech
The engineering achievement is impressive, but the economic viability is what makes the project truly disruptive. Commercial thermal surveillance setups require specialized, high-priced hardware. Puri built her prototype using accessible commercial components totaling around three hundred dollars.
By lowering the financial threshold to entry, small conservation organizations gain access to automated detection tools that were previously reserved for well-funded government agencies. This shift proves that solving complex ecological crises doesn't always demand corporate R&D budgets or billion-dollar venture funding. Sometimes it requires a fresh perspective and someone willing to question why existing tools cost so much.
For her work, Puri earned a twenty-five thousand dollar scholarship as a Davidson Fellow, cementing her place among young innovators turning raw technical curiosity into real-world utility. Technology works best when it tackles problems that actually matter.
