Future Tense

Can a Police Drone Recognize Your Face?

Sometimes it matters less if you’re being surveilled than if you feel like you are.

Protesters sit in the road, as seen from above.
A drone photo shows people gathered to protest over the death of George Floyd on June 2 in New York City. Lokman Vural Elibol/Anadolu Agency via Getty Images

Since the death of George Floyd on May 25, Americans have taken to the streets to peacefully protest in unprecedented numbers, calling for an end to our national culture of racism and police brutality. These protests have, on too many occasions, been met with violent force from police, who have been caught on camera using tear gas, pepper spray, rubber bullets, and other supposedly less-lethal weapons against unarmed and compliant people. Police around the country are also devoting considerable time and energy to collecting intelligence on protesters and protest movements, with methods ranging from monitoring social media posts to aerial surveillance—sometimes, with drones.

Police, military, and federal government forces have regularly flown surveillance helicopters and small, crewed surveillance aircraft over protest areas, capturing real-time video and photographs of protest movements. The New York Times found that by mid-June, the Department of Homeland Security had captured more than 270 hours of surveillance footage of protests from helicopters, airplanes, and drones, data that was shared with a digital network accessible by other federal agencies and by police departments. On May 29, U.S. Customs and Border Protection even flew a large military Predator drone 20,000 feet over protests in Minneapolis, allegedly at the request of still-unidentified “federal law enforcement partners.”

Some U.S. police agencies have also begun flying small, multirotor drones made by civilian manufacturers over protest marches and rallies, using the devices to capture real-time information on what’s happening on the ground. On social media, protesters have begun warning others to keep their masks on, afraid that authorities will use these drone-captured photos to identify protesters, using ever-more-common police facial recognition technologies. “Drones … represent the power of new surveillance tech as deployed by law enforcement—a new future where they have at their disposal incredibly powerful new devices,” said Jay Stanley, senior policy analyst at the ACLU.

The presence of drones at protests sets a worrisome precedent: Their deployment over even peaceful protests as a preemptive measure normalizes persistent surveillance and reminds protesters that they’re always being watched. But it’s important to clarify what police drones are, what they can and can’t do, and what their true purpose is when they hover over a protest.

While police drones are more common than they’ve ever been before (especially in the response to COVID-19, as I’ve recently written about), they aren’t exactly new. American law enforcement agencies have been experimenting with small drones for well over a decade, a trend that’s accelerated since the 2016 passage of the Federal Aviation Administration’s landmark “Part 107” rule, which allowed licensed nonhobbyists—including private businesses and government—to operate drones in U.S. airspace. March 2020 data from the Center for the Study of the Drone found that at least 1,578 state and local public safety agencies in the U.S. have drones, and 70 percent of these disclosed agencies are law enforcement.

However, these small police drones are very different beasts from the CBP’s high-altitude, thousand-pound Predator drone typically associated with American military use abroad. Most police drones today are consumer models produced by the same companies that make the drone your dad got for Christmas, like Chinese drone-maker DJI: These companies often market their aircraft to public safety agencies but don’t sell exclusively to them. Unlike a Predator—which is capable of staying aloft for more than a day—these small drones usually have short battery lives, from as little as 16 minutes, when carrying a very heavy camera, to 35 minutes when carrying a lighter sensor. (Drone evasion tip: If you think you’re being followed, duck under a shelter or a convenient tree. You can probably wait the drone’s battery out.)

Police drone users are largely not exempt from the same rules that other drone users must abide by, which include restrictions on flight over people, at night, and beyond the pilot’s “visual line of sight.” Beyond that, drone use by U.S. police is not standardized at the federal level and varies widely from agency and agency: Many state and local statutes place restrictions on what police can legally use small drones for, such as mandating that police obtain a warrant before they can collect drone data. Police drone users also need to get a special Part 107 waiver or a certificate of authorization from the FAA for usually off-limits actions like operating drones at night or over people—both activities we’ve seen a lot of during the past few weeks of protest. (These are published online, so you can even see whether your local law enforcement agencies hold waivers for these activities.)

So what do police use all these drones for? Mostly for situational awareness. Aerial drone shots give police a clearer sense of where crowds are moving and where vehicles are going, and can help them stay one step ahead of protesters. Police also use drones to find or follow people in certain scenarios, like filming setup drug deals, following burglary suspects on the run, or tracking down a suspect mired in a swamp. Some drones are now equipped with on-board sensors that can automatically follow a moving person or a vehicle, albeit under very limited conditions (and for a short period of time).

That’s probably unsettling, but here’s comforting news. While a police drone can certainly chase someone for a bit, that doesn’t mean police can readily use drone-collected imagery to identify who that person is. In my research for this piece, I couldn’t find a single example of U.S. law enforcement using facial recognition technology and drone imagery to identify someone in the real world. This almost certainly isn’t because police don’t want to, or because they’ve been legally barred from doing so. It’s because accurately recognizing individual people from aerial drone imagery is really, really hard.

Facial recognition technology in general still suffers from serious accuracy problems. While current verification algorithms can achieve very high levels of accuracy, that’s only the case under ideal environmental conditions: where lighting is perfect, the subject’s face is clear and totally unobscured, and (importantly) the camera is low enough to the ground and close enough to the subject to get a good picture. Under imperfect, “real-world” conditions (like, say, a protest where most people are wearing masks), errors become more common. Even the CEO of Clearview AI, the world’s creepiest facial recognition company, lamented to the New York Times that his app doesn’t work well when it’s given photographs taken from surveillance cameras that are placed “too high” on the wall or on the ceiling.

Real-world tests of facial recognition technology, conducted by independent researchers, have revealed disturbingly poor accuracy rates. Academic evaluators found that London’s Metropolitan Police facial recognition system had a whopping 81 percent error rate in real-world trials. Researchers have also discovered that common facial recognition techniques are far better at accurately identifying white males than they are at identifying people from other demographic groups, errors that could lead to more nonwhite people being falsely accused of crimes. This has already happened: In January, a Black man in Michigan was wrongfully arrested after a police facial recognition system matched his photo with that of a shoplifter.

If it’s hard to accurately identify faces from a camera on the ground that doesn’t go anywhere, we can assume that it’s even harder to identify faces from pictures taken by a moving drone, where the camera’s view of the person being surveilled is constantly changing. While many researchers (and the U.S. military) are certainly trying to figure out how to recognize people and faces from drone imagery, it appears to be slow going.

Although some scientists have had some success in applying person-recognition algorithms to drone imagery, they’ve only been able to accomplish this in highly controlled lab settings—and there are many remaining technical issues to resolve. While some companies and police forces say they have (or are planning to get) high-accuracy drone facial recognition technology, I couldn’t find any well-supported evidence of drone facial recognition technology being used to identify individual people in the wild. And humans, as it turns out, are even worse at identifying people from aerial photos than computers are. A 2017 study found that human analysts failed miserably at identifying individuals from aerial imagery shot by a low-flying drone. While it’s possible that researchers will figure out how to accurately ID individuals from aerial photographs in the future, the technology simply isn’t up to the task yet, and I suspect that it won’t be anytime soon. That’s a good thing, because it gives us time to figure out how to constrain it. There are still no national laws in the United States that limit police use of facial recognition, and while cities and states have developed their own policies, they are only partial solutions. We should demand that the federal government write laws that adequately protect people from facial recognition of all types, both in the air and on the ground.

But the point of drones may not be surveillance itself. As surveillance experts have long observed, from the pyramid builders of ancient Cahokia to Foucault, people who think they’re being watched at all times (even if they actually might not be) are much easier to frighten and to control. Jake Laperruque, senior counsel at the Project on Government Oversight, thinks that the mere presence of drones at peaceful protests could create a “chilling effect.” He told me, “Just the possibility [of being identified] can have a problematic effect for First Amendment expression and for protests.”

Disquieting as police drones are, you shouldn’t let their presence, and their symbolic powers of surveillance, intimidate you out of protesting, reporting, or other means of exercising your rights. While we should all be concerned about the potentially privacy-violating presence of small drones at protests, they’re something of a distraction from more powerful and effective law enforcement tools for identifying protesters, like smartphone tracking and surveillance cameras, which use technologies that are more proven and often less obvious than a buzzing, blinking drone.

Future Tense is a partnership of Slate, New America, and Arizona State University that examines emerging technologies, public policy, and society.