Pentagon testing mass surveillance balloons across the US (2019) | Hacker News Reader

Pentagon testing mass surveillance balloons across the US (2019)

theguardian.com

56 pointsTheFreim22 comments

These freedom balloons or whatever, they have a neat trick I remember reading about. In principle, if you're at very high altitude, you can simultaneously image everything over hundreds of square miles, with the right optics and an extremely large camera array (many gigapixels). You're close enough to identify faces (unlike orbiting satellites 200 km+ up). And you can stare at everything forever, 24/7, except that the firehose of data is unmanageable (even just a direct downlink would be impossible). So the solution, the trick, is a large FPGA on the balloon that classifies small pixel regions, according to whether "these pixels are static"; "these pixels might contain a human"; or "these pixels are non-static but probably not human". And the result is to select small windows out of the firehose for real-time downlink, while discarding the rest (or just updating it at low refresh rate).

I don't know if this was common knowledge or not: but the effect is you can continuously stare at incredibly large swaths of land at once, with the illusion of a pseudo-100+ Gbps video feed, through the right combination of vehicle/optics/camera/signal processing.

You're close enough to identify faces

I seriously doubt that even with black-projects AI assistance. I have no experience with hi-altitude fancy photography, but a lot with the limits of terrestrial photography of human subjects.

I work on aerial imaging hardware and… I’ll bite, just for a few minutes. At this point I’m at 50/50 and just want to know the answer at “napkin resolution”

https://www.axis.com/learning/web-articles/identification-an... Suggests something like a 2mm-4mm GSD would be required.

The small city I live in has an area of 179km^2. Assuming it’s a square and we’re taking a square photo, that’s 13.37km x 13.37km (nice number!). In each axis this is 13.37e3 m / 2e-3 m/px = 66.85e6 pixels. Squaring that gives us… if I’m counting right, a 4.4 Exapixel image. If we used the 4mm GSD instead of the 2mm GSD it would still be a 1.1 Exapixel image.

On a modest budget (maybe $200-300k) the kit I work with could be scaled up to do real-time gigapixel imaging at maybe 20-30 fps. But that’s still short by a factor of 1M both in image sensor real estate and in processing capacity.

I agree that this seems implausible. That being said, the gigapixel rig could still be quite scary in the wrong hands, just not at the city scale. You could do video that covers about a 350ft x 350ft area and be able to retroactively go back and do face identification on everyone whose faces you can see. It’s a completely parallel problem too and the costs scale linearly with the area unless you start running into second-order effects like diffraction limits or obliquely angles.

Edit: option 2 now that I think about it a little more would be to have a lower resolution broad image with servoable high resolution units. Point the 350x350ft camera at a specific ROI and filter down from there. That $200-300k budget includes about a dozen onboard high performance ML engines. Depending on the actual frame rate you want processed, you could either do light object recognition at a reasonable rate, or heavy inference at maybe 0.5-1fps.

Looks correct to me (except 4e15 is a peta- rather an exa-, but that's tangential and doesn't seem to affect your real conclusion, the "factor of 1M"). The "facial recognition" must be me misremembering the original writeup, which I still haven't found.

It's probably lame to double down on this point, but, gigapixels is definitely enough to contiguously track people back to their homes and workplaces, which gives you like 90% of the benefit of facial recognition (for, as you point out, 1e-6 of the cost).

Thanks for the correction! I was in the tub and not really in the "SI units" headspace :)

And yeah, continuous tracking of people is definitely viable, including entire crowds of people. A couple of years ago our local football team did a gigapixel photo of the entire stadium (on a 360 degree pan mount) and you could very easily manually pan through the photo and identify faces very very clearly. Tech to do that in real-time is definitely here and readily available.

Yes, but notice how they did not test them in Chinese airspace.

Didn't they?

No, the Chinese get aggressive if anyone flies in the South China Sea let alone over China proper.

Exactly, China was boasting about it's ability to shoot down high altitude balloons in 2019. You can still find some stuff in Mandarin, but everything in english, has been taken down. They were boasting about their high altitude surveillance balloons a year ago, and again you can still find some original material, but all of the english language articles have been removed. The propaganda machine is spinning on it's head now.

Which is neither here, nor there

What would be the purpose of surveillance balloons if not to use them on your rivals...

It is cost benefit trade off. Sure US could get some benefit but at the expense of the similar scenario that is playing out now here. Using balloons over national territory is highly escalatory. Whatever they're getting intelligence wise it can't be worth the fallout. Building a capability is not the same as employing it. We have nukes and yet we don't regularly bring them out.

To surveil your population, as most "first world" governments seem so keen on?

When the citizenry has the ability to remove you from power, they become your rivals

They use it on rivals. One of the rivals is their own people.

Notice that we have zero evidence that the recent objects were Chinese in origin except the admin has said so. And we know they never lie about anything.

The Chinese said it was their balloon last week. They want it back. If we just go with the prevailing trend, China is the most likely explanation for these by a long shot.

If China lied about its balloons (over Beijing) being from the US, would the US attempt to do the impossible and prove that something never existed or agree and ask for them back?

Come to think of it, if the Chinese want this balloon back, does it make it more or less likely that this is a US balloon?

(then again, if it is, presumably the US wouldnt simply hand them back "as is", so not sure what kind of endgame that would serve)

It would be funny if they turned out to be ours. A case of the right hand not knowing what the left hand was doing.

the first one they got had hardware from several Chinese corps and had every sign of being from China, are you denying that? I guarantee if they got even 2% of the hardware they were able to figure out where it came from.

> Notice that we have zero evidence that the recent objects were Chinese in origin

It's written right there. Made in China. The only thing still made in US is military equipment.

Reply on news.ycombinator.com