Army Developing Drones That Can Recognise Your Face From a Distance
Clay Dillow
at 07:01 AM 29 Sep 2020
Comments 0
It's not your friend
It's not your friend
IMAGE BY Complex
Military // 

It's not enough for the US military to be able to monitor you from afar. The US Army wants its drones to know you through and through, reports Danger Room, and it is imbuing them with the ability to recognise you in a crowd and even to know what you are thinking and feeling. Like a best friend that at any moment might vaporise you with a hellfire missile.

Of a handful of contracts just handed out by the Army, two are notable for their unique Intelligence, Surveillance and Reconnaissance (ISR) capabilities. One would arm drones with facial recognition software that can remember faces so targets can't disappear into crowds. The other sounds far more unsettling: a human behaviour engine capable of stacking informant info against intelligence data against other evidence to predict a person's intent. That's right: the act of determining whether you are friend or foe could be turned over to the machines.

That's a bit disquieting whether you are an insurgent warfighter or not. But back to the overarching topic at hand: The US military is pulling in more ISR data than it knows what to do with these days, a lot of it useless noise that's inconsequential to ongoing operations. And, as Danger Room notes, the strategy in Afghanistan has changed from one of winning hearts and minds through nation-building projects to targeting specific bad guys.

The hard part is keeping up with said bad guys, and that's where Progeny Systems Corporation's "Long Range, Non-cooperative, Biometric Tagging, Tracking and Location" system comes into play. The facial recognition layer of its technology is pretty standard: take some 2D pictures of a target's face, use them to build a 3D model, and then use that 3D model to recognise the face later.

But that's not necessarily easy. It's difficult enough for computers to pull off biometric facial recognition when the subject is stationary and looking straight at the camera. Toss in the many variables inherent in aerial ISR - a moving target who may be in profile or looking downward, a moving drone, low resolution cameras - and it's a major challenge.

Progeny's system, if it works the way the company and the Army envision it, needs just 50 pixels between the target's eyes in a 2D image to build the 3D model. "Any pose, any expression, any face," the company's lead biometric researcher tells Danger Room. From that model stored in Progeny's database, the system could identify the target from an even lower resolution image or video.

The closer the drone is to the subject, the better all of this works. But progeny also layers in a second kind of recognition that can work at more than 230 metres. This "soft biometric" system basically takes in a bunch of non-facial but otherwise outwardly relevant data - skin colour, height and build, age, gender - to build a larger kind of model for its vision algorithms to work with. If a body is moving through the crowd, Progeny claims that a drone circling high overhead can keep track of him or her simply using this larger, whole-body identification system.

But what good is tracking if you don't know who your enemies are? Another contract handed out to Charles River Analytics seeks to develop a human behaviour engine known as Adversary Behaviour Acquisition, Collection, Understanding, and Summarisation (ABACUS). It mashes up all kinds of behavioural data into a system that churns out an assessment of adversarial intent, determining if a subject has enough built up resentment toward the US and its aims to be a potential threat.

So pretty soon the drones may know who you are, where you're going, and what you're planning to do when you get there.

[Danger Room]

 
0 COMMENTS

Leave a comment

Please provide your details to leave a comment.

The fields marked with (*) are required.


Display Name: *
Email *:
Comments *:
(Max 1000 characters)
 
BY Danika Wilkinson POSTED 12.10.2020 | 0 COMMENTS
BY Danika Wilkinson POSTED 12.10.2020 | 0 COMMENTS
BY Nick Gilbert POSTED 07.10.2020 | 1 COMMENT