The Military Is Building Long-Range Facial Recognition That Works in the Dark (medium.com) 60
An anonymous reader shares a report: The U.S. military is spending more than $4.5 million to develop facial recognition technology that reads the pattern of heat being emitted by faces in order to identify specific people. The technology would work in the dark and across long distances, according to contracts posted on a federal spending database. Facial recognition is already employed by the military, which uses the technology to identify individuals on the battlefield. But existing facial recognition technology typically relies on images generated by standard cameras, such as those found in iPhone or CCTV networks.
Now, the military wants to develop a facial recognition system that analyzes infrared images to identify individuals. The Army Research Lab has previously publicized research in this area, but these contracts, which started at the end of September 2019 and run until 2021, indicate the technology is now being actively developed for use in the field. "Sensors should be demonstrable in environments such as targets seen through automotive windshield glass, targets that are backlit, and targets that are obscured due to light weather (e.g., fog)," the Department of Defense indicated when requesting proposals.
Now, the military wants to develop a facial recognition system that analyzes infrared images to identify individuals. The Army Research Lab has previously publicized research in this area, but these contracts, which started at the end of September 2019 and run until 2021, indicate the technology is now being actively developed for use in the field. "Sensors should be demonstrable in environments such as targets seen through automotive windshield glass, targets that are backlit, and targets that are obscured due to light weather (e.g., fog)," the Department of Defense indicated when requesting proposals.
Re: (Score:1)
Heat signatures are different from visual recognition, the failings with regard to people of color is due to lack of sufficient contrast and abilities of cameras, coupled with how the related AI is trained.
Re: (Score:1)
Re: (Score:2)
I remember in 5th grade when my best friend came back from a family vacation in Mexico with a bad sunburn and I was like, "Woah." And he was like, "Yeah, the teacher was wrong."
Re: (Score:2)
The can't even recognize black people reliably nearby in pain sunlight.
Well, they don't need to - they just shoot them anyway; they probably have ID in their wallets.
Re: (Score:2)
That's cops.
This is the military.
There is usually nothing left in the pockets. Including the pocket.
Re: (Score:2)
No, dumb ass, that's exactly the issue that makes it so that when I see "that works in the dark," I think, "oh shit, somebody actually got serious about solving this problem!"
If it works in the dark, it isn't using visible light, and so none of the (obvious and predicted) problems with cheapskate facial recognition will be guaranteed to be present in this tech.
Re: (Score:2)
The can't even recognize black people reliably nearby in pain sunlight.
Seriously, watch Better Off Ted [wikipedia.org], Season 1, Episode 4, Racial Sensitivity [fandom.com].
Veridian Dynamics [veridian-dynamics.org] installs new motion sensors in the building, which detect employees based on the light reflecting off their skin. In short, they can't detect black people -- leaving them stranded in offices, the elevator, and not being able to use the water fountain, etc... (almost everything in the building is motion-activated).
Rather than admit a mistake and revert to the old system, Veridian tries to continue on with a few si
Define (Score:5, Funny)
Long range:
The device should be able to operate from a distance of 10 to 500 meters and match individuals against a watchlist.
So they want a sensor that can work through a car windshield and detect subtle heat differences over 500 meters in sufficient detail to identify certain individuals?
Seems unlikely.
Re: (Score:2)
Do thermal sensors even work through glass?
Re:Define (Score:5, Insightful)
Do thermal sensors even work through glass?
https://infraredtraininginstit... [slashdot.org]>Nope. Nor do they work in fog. If this was a promising way to differentiate otherwise identical looking faces the commercial value alone would far exceed 4.5 million. It reads like a personal handout to a pre-existing company disguised as a contract proposal.
Re: (Score:2)
Re: (Score:2)
Cluestick, that link doesn't say what you said, or support your claim.
You have to know how sensitive the technology is in open air before you can apply the numbers in that link to determine how the glass will affect the range. Duh.
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
Because cameras like to maximize transmissiveness, not because it is black and white and it is 100% blocked. (lol)
Re: (Score:1)
Yes, they work fine. The glass dampens the signal, but the technology of course doesn't care about the absolute sensor levels, it cares about relative levels from one pixel to the next.
The world isn't black and white, transmission of photons of different wavelenths do not all get blocked or all pass through a thin substance like window glass.
Re: (Score:2)
Re: (Score:2)
I'm assuming you replied to the wrong post, since you're repeating what I said with less explanation.
Re: (Score:2)
I'm assuming you replied to the wrong post, since you're repeating what I said with less explanation.
You realize 5g doesn't even work through windows [iotforall.com] right?
Re: (Score:2)
I said it, I repeated it by implication, and now again.
You're about as literate as a tree frog.
Re: (Score:2)
Yes, they work fine. The glass dampens the signal, but the technology of course doesn't care about the absolute sensor levels, it cares about relative levels from one pixel to the next.
The world isn't black and white, transmission of photons of different wavelenths do not all get blocked or all pass through a thin substance like window glass.
No, they (thermal sensors through silicon dioxide) do not work fine. For any real sensor in the real world, yes *all* incoming photons are blocked and there is no discernible difference between pixels outside of noise. In fact, they are absorbed and on a smooth surface like a windshield, roughly 30% are reflected just like a mirror. Head on, the only image you would capture is a faint one of yourself. When you are factually correct, you don’t need to resort to ad hominem.
Re: (Score:2)
Dear dumb ass.
Learn yous tu reed
Re: (Score:2)
Yes, they work fine. The glass dampens the signal, but the technology of course doesn't care about the absolute sensor levels, it cares about relative levels from one pixel to the next.
I have a far infrared camera, a real one with a cooled sensor, and it does not see anything through glass. Glass actually makes a fine mirror for far infrared and when I have tested it, what I saw was the far infrared reflection while visible light passed right through as expected. It was a little weird holding a glass plate up and seeing only a head bobbing along with feet below and a trail of footprints.
Re: (Score:2)
You probably haven't tested it well enough to make those claims, or you're just lying.
You're most likely conflating "I can't make anything out in the resulting image" with "nothing was recorded on the sensor." But those are very different.
If you have any sort of scientific background you'd already know that something like this will never be 100%, it is.... stupid... to think it does. It blocks some percentage of the light at a certain frequency. Enough that when the image sensor is calibrated for normal ima
Re: (Score:2)
4.5 million probably covers a Flir camera that has enough lensing and resolution to determine a blob is a human at 500m.
Re: (Score:3)
Long range:
The device should be able to operate from a distance of 10 to 500 meters and match individuals against a watchlist.
So they want a sensor that can work through a car windshield and detect subtle heat differences over 500 meters in sufficient detail to identify certain individuals?
Seems unlikely.
It is the US military. They do not care if they have to murder 10 civilians in order to assassinate one target. They have demonstrated that repeatedly. Hence accuracy does not need to be that great.
Re: (Score:1)
Murder is, by definition, unlawful.
I understand you want to express your opinion, but instead of doing that you chose to make an objective lie.
Even the root of the word murder originally meant "killing at night." From the earliest concept of "murder" it never applied to lawful killing, such as that done on a field of battle according to contemporary ethical principles of warfare. Just like, it doesn't apply to accidental killings that could or should have been prevented.
Re: (Score:2)
And then you look at the checklist for murder and you find it matches. The target of the assassination may, by some convoluted reasoning, even be a legitimate target for a legal kill, but even that is doubtful. The bystanders are not, unless there is a hot, declared war going on. Incidentally, only rogue-nations (vulgo: Criminal States) do not respect the Geneva Convention.
The lie is entirely on your side here. But I suspect you are not smart enough to see that and are just regurgitating propaganda.
Re: (Score:2)
No, you're not looking at a checklist for murder, you're simply engaging in non-literal hyperbole. Also known as regurgitating propaganda.
Because you found it morally objectionable does mean that you think it is murder. It only means you find it morally objectionable, also known as "wrong." So you could call it a "wrongful killing" and you'd be expressing your opinion. When you call it murder, you're making a literal and false accusation which is clearly morally reprehensible.
Re: (Score:2)
It is the US military. They do not care if they have to murder 10 civilians in order to assassinate one target. They have demonstrated that repeatedly. Hence accuracy does not need to be that great.
The US military traditionally tries to minimize collateral.
In WWII it took boxes and boxes of bombers (think 1000's of planes) to take out one ball-bearing factory.. miss, and wipe out the school and hospital instead.
By Vietnam it was the beginning of precision-guided munitions.
By Gulf War I it was mature, but now we have JDAM, much better and impervious to weather.
Don't we have a knife-bomb now too?
Re: (Score:2)
The US military traditionally tries to minimize collateral.
"Tries to" is the key concept. But it becomes a war crime (or because there is no declared war, just a crime) when they do the strike well knowing there will significant non-target victims. And they do the strike, even when sure of that. That they reduce the numbers is just changed tech and economics of military strikes (drones are used because they are _cheap_ in comparison to the alternatives), more public scrutiny and the realization that you can win on the military side but if you murder too many pregna
Re: (Score:1)
Re: (Score:2)
So they want a sensor that can work through a car windshield and detect subtle heat differences over 500 meters in sufficient detail to identify certain individuals?
Seems unlikely.
It is doubly so unlikely since far infrared does not pass through glass.
$4.5 million? (Score:2)
I'll do it for $1 million.
"standard cameras, such as those found in iPhone"
Whut?
Re: (Score:2)
Re: (Score:2)
You have the business smarts. We should start a "thinktank" together.
Re: (Score:1)
Will it work when it's not dark? (Score:2)
Will it work when it's not dark? Sometimes specifications tend to omit the obvious.
Re: (Score:2)
Yes, but not if the target is wearing sunglasses in the dark.
Training dataset (Score:4, Interesting)
Where will they get the biometric data necessary to make identifications? There exists plenty of visible light image sources for standard facial recognition. But it's my understanding that the IR signature of one's face differs enough from the visible to make a conversion difficult.
IR facial recognition is in use today for high security facilities. For one thing, it makes holding a picture of a subject's face in front of a camera useless. But that depends on each individual's IR facial photograph being captured. Not a problem when you are sitting for a badge photograph. I suspect that the fancy camera with multiple lenses was doing exactly this when I had my photo taken.
Re: (Score:2)
Many airports have deployed IR cameras for years.
Re: (Score:2)
They don't choose targets off a list of names, though. By the time they know they want to kill somebody, they've already surveilled them.
They will use the same technology to record the data in the first place; some human asset who has access to the person will take a selfie with them or something, and now they have the data. That's in the hard cases.
In most cases, they know who the militants are, and were already tracking them when they started interacting with other people who are already radicalized. They
Re: (Score:2)
Re: (Score:2)
From their database of military members and employees.
They have no interest in identifying "military members and employees" from 10 - 500 meters away The DoD is calling for the technology to be incorporated into a device that is small enough to be carried by an individual. The device should be able to operate from a distance of 10 to 500 meters and match individuals against a watchlist.
This tech would be useful at base gates, secure area, building entrances, etc. to spot people who are not authorized.
See above, this is designed to work from a "watchlist", not an "everybody in the world I already know list" - the storage requirements are much simpler for the "watchlist" mode
Re: (Score:2)
Re: (Score:2)
The RFP specifies how the device will be used:
According to the DoD request, the new device will be used to identify those on a watch list rather than combing through the DFBA’s entire database.
The request also states it should be portable, not fixed:
The DoD is calling for the technology to be incorporated into a device that is small enough to be carried by an individual.
The request also states it is intended to operate from a watchlist, not to point out people that aren't on the watch list.
The device should be able to operate from a distance of 10 to 500 meters and match individuals against a watchlist.
So this isn't a request to identify who's allowed on a base - that has been pretty much Sussed out - this is clearly something different than you described, dipshit.
Eeee Gads! (Score:2)
I hope I don't have a doppelganger that is a terrorist.
Or is banging that military guys wife.....
Re: (Score:2)
More importantly, don't travel to a war zone, or you might have an unplanned explosion. Your concerns are ridiculously marginal compared to the baseline dangers.
China does the same (Score:2)
Re: (Score:2)
Xi Jinping: ( dum de dum time to go through the lavatory screening, gotta go, gotta go right now ) Hi there LavBot, let me in.
LavBot: Ummm....I don't know, you look an awful lot like...wait a minute...gotta check my records...hmmm...sez here you are Ms. Xi Jinping. This is the man's lavatory.
Jinping: Nooooo...look at here, see, I'm Mr. Xi Jinping.
LavBot: We require facial recognition and don't show me that thing again, it's embarrassing.
Jinping: (now doing the pee-pee dance in double time) Damn it, I'm the
Pointless (Score:3)
Assuming that the scheme is even workable, which I doubt, it would be trivial to camouflage or disguise against it. It would just lead to more men wearing makeup.
Re: (Score:2)
What if that makes them stand out more as militants?
It is easy to defeat the technology if you're willing to make yourself stand out more, just wear a Guy Fawkes mask. Works great in war zones when you're approaching a road block or embassy, or just passing a military convoy.
Re: (Score:2)
This device is thermal, just be sure and keep warm or cold things in front of your face - for example, your hand, a ski mask, or a kerchief/keffiyeh - remember, this is for operating at night, in the fog, etc - don't worry about looking silly.
My tin hat is useless (Score:1)
Now I need a tin mask, insulated to block the IR I guess
That should make a real fashion statement
Re: (Score:2)
A ski mask would likely alter your thermal image for the sensor, as would simply holding your hand over your face.
Of course, then you can just start taking out anyone walking around in the dark with their hands over their faces...
Infrared (Score:4, Interesting)
"Sensors should be demonstrable in environments such as targets seen through automotive windshield glass,"
If the driver has the heat full up to the windshield, the glass should be much warmer than his face, I'm not sure it would work then.