An Israeli biometrics startup called AnyVision with ties to Israel’s military has applied for a U.S. patent on technology that tells drones how to maneuver to capture better facial recognition images of people on the ground.
Facial recognition technology has become widely used by law enforcement around the world, but the technology is controversial in part for its accuracy issues, especially when recognizing Black and brown faces. Activists are now calling for ending its use entirely, and police use of facial recognition has already been banned in a host of U.S. cities.
The patent application, titled “Adaptive Positioning of Drones for Enhanced Face Recognition,” describes a computer vision system that analyzes the angle of a drone camera in relation to the face of a person on the ground, then instructs the drone on how to improve its vantage point. The system can then send that image through a machine-learning model trained to classify individual faces. The model sends back a classification with a probability score. If the probability score falls below a certain threshold, the whole process starts over again.
A future defined by this type of mass surveillance would “obliterate privacy and anonymity in public as we know it,” said Kade Crockford, head of the Technology for Liberty Program at the ACLU of Massachusetts who’s led the charge on banning facial recognition in Massachusetts cities, in an interview with Fast Company last year. “Weirdly this is not a hugely controversial issue for voters. People don’t want the government to be tracking them by their face every time they leave their house.”
People don’t want the government to be tracking them by their face every time they leave their house.”
As with any patent application, there’s no guarantee the technology will show up in a real product. But it does address a very real technical problem with existing facial recognition systems. Such systems usually process images captured by stationary cameras. Capturing a clear angle on someone’s face, and compensating for bad ones, is always a challenge with these systems. Shooting video from drones that can move around and intelligently zero in on the right angle is a way of taking the chance out of the process.
The application, which was originally reported by Forbes cybersecurity writer Thomas Brewster, was filed last summer and published by the U.S. Patent Office on February 4.
AnyVision, which was founded in 2015, sells artificial intelligence designed to let cameras in retail stores recognize the faces of people on “watch lists” who have been convicted of theft in the past. The technology can also support contactless entry systems where a person’s face acts as their “key” to go through a door or past a turnstile.
“Facial recognition with drones is a technology that may be used in the future for package delivery,” AnyVision CEO Avi Golan said in an email statement to Fast Company. “Any major player in the delivery business is looking at ‘last mile’ solutions including facial recognition for fast and easy personal identification.” Golan says drone facial recognition technology might also be used in mines to keep track of employees for safety purposes.
“AnyVision is not involved in weapons development and is focused on the many opportunities in the civilian market,” Golan wrote.
But the company’s technology is being used in defense applications, for security. AnyVision found itself in the middle of a controversy when Israeli daily Haaretz reported in June 2019 that its technology was being used by the Israeli military in a secret surveillance program to recognize Palestinian faces “deep inside the West Bank.” The company insisted that its technology is used only at border crossings.
At the time, Microsoft, which was a minority investor in AnyVision, contracted a legal team led by former U.S. Attorney General Eric Holder to conduct an independent audit of the startup and the claims. Holder’s team found the allegations to be false. But very soon afterward, in March 2020, Microsoft divested its stake in AnyVision, announcing that it would no longer invest in facial recognition startups. DFJ Growth, Qualcomm Ventures, and Lightspeed Venture Partners have also invested in AnyVision, according to Crunchbase.
That was two months before the May 2020 murder of George Floyd, which prompted Microsoft and a number of Big Tech players to either temporarily or permanently stop selling their own facial recognition AI to police departments. Facial recognition technology has been shown to misidentify, or falsely match, Black and brown faces in particular, contributing to systemic racism within policing. A Georgetown Law School study found that more than half of local police departments in the U.S. already use the technology.
Even though several tech giants have stepped away from selling facial recognition to law enforcement, a wave of smaller companies like AnyVision have been quietly but aggressively pursuing contracts with police, military, institutions (such as hospitals), and retailers. Biometrics is a quickly growing business, and its growth has been further accelerated during the pandemic as contactless identification has become very important. The research firm Markets and Markets (that’s really the name) reported in late 2020 that sales of biometric systems will nearly double from $36.6 billion in 2020 to $68.6 billion in 2025.
AnyVision is vocal about the bias problem, and says its system proved to be more than 99% accurate during a public challenge of 150 facial recognition algorithms to evaluate accuracy in detecting gender and skin color.
However, even if the system is accurate, critics say that it continues to push us toward a future of mass surveillance and could have a chilling effect on legitimate dissent and protest—especially when there is no transparency or accountability about how facial recognition systems are built, who gets to use them, and for what purpose.
And after all, watching for suspected terrorists at the border is one thing, but it’s not hard to imagine AnyVision’s positioning system being used for drones that aim more than just cameras.