[nine_k]

Numbers don't add up.

If you take a 8K camera with a standard 50 mm lens, its angular resolution is about 20" / pixel.

A 50 mm lens has a FOV of about 40°. It covers a cone of about 0.38 strad. A full hemisphere has 2·pi = 6.28 strad, so we need at least 16.5 such cones to cover the whole area; actually we need likely 20-25 because of imperfect geometry and some safety margins at intersections. We can, of course, mount fewer and scan.

If we take a plane like A320 (larger than a typical fighter jet), and remove it 25 km from us, its angular size would be about about 5', or 300". Our A320 would be 15 pixels wide, assuming very good optics, and very clear skies. This is not much to determine what craft is approaching us. At the cruise speed of 800 km/h, or 220 m/s, the plane will reach us in 122 s, or less than 2 minutes. Not a lot of warning. A fighter jet making 500 m/s would be there in 50 s.

This is, of course, without any clouds. Even very light clouds or haze would conceal the aircraft at 25 km. To say nothing of the night time.

We could of course take in IR camera, but I don't remember 8K IR cameras being cheap, or even available. A stealth aircraft like B-2 does a lot to make its thermal signature very faint, including the exhaust.

[MindSpunk]

IR sensor based tracking has existed for decades [1]. But aircraft equipped with them still equipped radar systems because of their significantly longer range tracking capability. Optical has all the same downsides as IR tracking, on top of not working at night. It's like people forget that radar relies on the scanning aircraft to actively illuminate their targets with the radar emitter, where as optical systems rely on the sun to do that. Targets are at least partially self illuminating for IR trackers thanks to the hot jet engines.

[1] https://en.wikipedia.org/wiki/Infrared_search_and_track

[aeternum]

Most camera sensors already detect in the IR range, you just need to remove the filter. I've done it, works great and pretty fun to convert some old webcams, robot vacuums lidar scanners look especially cool.

Blurry video is fine when using techniques like this: https://www.youtube.com/watch?v=m-b51C82-UE

Clouds also don't save you (unless you have two thick layers to fly through) because this technique is even easier with satellites. Stealth effectively no longer exists for most nation-state level tech. The B-2 is a very cool plane but is unfortunately obsolete. Still great for when you want to put on more of a show than an attack.

[Tor3]

"Clouds also don't save you (unless you have two thick layers to fly through) because this technique is even easier with satellites."

This is incorrect. A typical satellite will orbit once every 100 minutes or so (military spy satellites more often because they fly lower, but that only makes the next part even worse). To have any kind of resolution the swath it can scan is very narrow. It'll pass from horizon to horizon in some 10-14 minutes or so, if if passes reasonably overhead (which it'll do once, the next orbit it'll be far from overhead or not seen at all, depending on your latitude).

For a satellite to spot an airplane you need to be in luck. A coincidence. It's not something you can use for spotting airplanes. The harder you look (the more you increase resolution) the more narrow the swath gets. You can have more satellites. There's still no chance of actively detecting airplanes on a regular basis. And this doesn't even take into consideration that the data must be processed after having been dumped from the satellite. The satellite is by then elsewhere.

You could use a geostationary satellite, to monitor a good third of the planet at once. But then you're nearly 36000km above equator and you can't see any details. So, not that either.

Satellites are great for scanning the surface of the planet. And for that we're now at a stage where it's hard to hide anything, for very long at least. But moving airplanes is something entirely different.

(My job is about processing data from satellites).

[overfeed]

> A typical satellite will orbit once every 100 minutes or so (military spy satellites more often because they fly lower,

Constellations (like Star Shield or the Chinese equivalent) solve this problem; there is always a dozens of satellites overhead, and they don't need a lot of resolving power to detect contrails, I vet even cubesats with repurposed phone-camera sensors would suffice.

[Tor3]

contrails can be detected, at least when they disperse a bit. And as they linger it's even easier. What's difficult anyway you look at it is to detect the actual airplane, at the point where it's at at the moment.

Note that the Starshield satellites aren't able to cover all of the earth all of the time, far from it.

[overfeed]

> What's difficult anyway you look at it is to detect the actual airplane,

Contrails tell you when and where to look with more paranoia than usual, and temporarily lower the noise floor and not to disregard the some signals that you'd normally classify as a small bird.

[myself248]

> Most camera sensors already detect in the IR range, you just need to remove the filter.

Yes, but no. "Infrared" is a very wide range, and you're talking about different things here.

"Just off the red end of the visible spectrum", aka near-IR is typically considered 700-1400nm, which is what your normal visible-spectrum camera becomes sensitive to when you pop off the filter. That's fun, and you'll find lots of cool things there. Remote controls use near-IR typically in 850nm, flowers often reflect vividly in this range too. Notably, near-IR passes through most materials that're clear in the visible spectrum, which is how bugs are able to have eyes that see it to locate those flowers. Also, plastics and glasses, so NIR-capable optics are cheap. (NIR windows are often dyed to look dark-purple or black to the visible spectrum, because these dyes are transparent to IR.)

However, "thermal infrared" is much longer -- rocket exhaust can be seen with a mediumwave-IR sensor in the 3000-8000nm range, but warm bodies only start to show up in longwave-IR, 8000-15000nm. Sensors for those are mindbogglingly harder to make than near-IR. And these wavelengths don't pass through normal materials. Plain old glass, for instance, is totally opaque to thermal wavelengths, so if you take a thermal picture (a thermograph) of a window, you don't see the warm bodies inside, but rather the temperature of the glass itself, combined with whatever outside objects are reflected off its surface -- it acts like a mirror, not a window.

This means making lenses for thermal cameras is also difficult and expensive. The materials are awful -- Zinc Selenide is one of the most common, despite being expensive and toxic so it's difficult to machine. Pure Germanium works, but it's even more expensive. Sodium Chloride is amusingly transparent to LWIR but it tends to dislike getting rained on.

Removing the hot-mirror from a visible-light camera is a neat party trick and does legitimately see into "the infrared", but that's not the same as thermal infrared, which does still require specialized equipment.

[aliher1911]

Curious if you saw any difference in far IR vs near IR on cameras with removed filters. When I tried camera without filter or camera with IR filter (one that only lets IR through and look redish black to naked eye) attached I had nice b/w images of trees on long exposures or better pictures of tv remote leds. But if I make a picture of an iron or pan they look the same when either cold or hot. This is very different from thermal cameras I used where you can clearly see unevennes of the pan heat, dark image on tshirt being warmer on the sun, person at night etc. I can speculate that thermal exhaust falls into the second category.

[lm28469]

> Most camera sensors already detect in the IR range, you just need to remove the filter. I've done it, works great and pretty fun

It's not the same IR you'd need for military purposes, you're seeing a tiny percentage of the IR spectrum, and it's the most useless. That's why any semi decent military tier system costs 50k+ a pop. Civilian tier scopes are 1k+ and pretty shit, you won't see anything smaller than a human more than 1km away.

[nine_k]

Does a consumer-grade 8K camera sensor detect a 20°C difference in a few pixels? Cameras have to aggressively fight against thermal noise. Visible light is much more energetic than IR; detecting a faint IR signal with a sensor which is about as hot as the object it's looking at is hard. You need to cool the sensor down; old military aircraft even carried a tank of liquid nitrogen for that.

I mean, yes, you can take a low-noise sensor, add cooling, add a telescope lens so that you'd see the shape more readily, put a bunch of these telescopes on a rotating platform to scan the sky, etc. This is doable, but the thread started with an idea that it's doable with consumer-grade ("cheap") tech. I doubt that.

While at it, even if we assume that stealth does not exist for fast and heavy aircraft, it seems to effectively exist for slow, lighter-weight drones. Ukrainian drones, built from ultralight aircraft like Aeroprakt A-20, somehow penetrate 700 miles into Russian territory to burn refineries. With a cruise speed of 70 mph (sic), it should take them 10 hours to fly this distance. Were they detected efficiently, that would be enough time to scramble an interceptor a hundred times. Apparently this does not happen.

[aeternum]

How many of those attacks were successful? Russia has fairly advanced radar that is quite capable of picking up a plane like that. I think the more likely explanation is that it was incorrectly deemed not a threat or a minor threat.

700 miles is far more than the standard range of the A-20 (210nm). Is it possible they launched it from well within Russia thereby making it much less likely to be considered a threat?

[joha4270]

Possible, just like Russel's teapot.

But there has been one very well published Ukrainian attack launched from Russian soil[1].

If Ukraine had could regularly smuggle several drone aircraft's and explosives into Russia and launch them, we would be seeing a lot of other effects. Such as attacks on weapon production sites deep into Russia[2].

But yes, a lot of those attacks are probably unsuccessful. With my qualifications as armchair general, I would be surprised if more than 10% of them was successful.

[1]: https://en.wikipedia.org/wiki/Operation_Spiderweb [2]: https://en.wikipedia.org/wiki/Yelabuga_drone_factory

[timcobb]

I also don't think they'll target the Yelabuga drone factory with a drone when they could use that drone on something relatively small but high value (like an aircraft) or combustible (like a refinery), and are waiting for their heavy cruise missles to come online for targets like factories. You can't do much damage to a factory with a drone. We've seen Ukraine target industrial sites with drones, but it's not common.

[nine_k]

These attacks were so successful that Russia is having fuel shortages right now, and decreased its exports of oil products by about 10%. It is estimated that about 30% of Russia's refining capacity is disabled by now. (E.g. https://www.bbc.com/news/articles/czx020k4056o)

So no, these are not isolated incidents, this is a big hole in air defense, and Russia used to produce some competent air defense systems. They are just not geared towards the drones, much like Patriot air defense systems are not either.

[aeternum]

I agree the attacks were successful, I just don't agree they were successful because of lack of detection.

You generally can't continually launch a even $1mil missiles at $100-$10k drones and expect to win. The problem is that most air defenses were designed in an age where the enemy aircraft were quite expensive so it wasn't so critical to optimize ordinance cost. That IMO is the primary challenge for all militaries now.

[nine_k]

You don't need a missile. Any fighter jet is equipped with a cannon; a bunch of 30mm fragmentation-action ammo would dispatch the drone. You don't need a jet; any decent piston-engine trainer plane can fly 4x as fast as the drone, just mount a machine gun on it. Hell, take a civilian helicopter, shoot through the open door. You have ten hours to dispatch it, given that you have detected it.

But, since this is not being done routinely along the way, I conclude that detection fails. All reports about drones shot down mention that the drones were shot down nearby, likely by local air-defense teams near the target. These could as well detect the drone visually, or by the sound.

[6mian]

In visible range, would. The technique shown in the video work not perfectly clear sky conditions? Clouds are not uniform layer that moves with exactly the same speed in each point.

Are there reliable ways to figure out which pixels on subtracted frames are cloud movement and which are the few ones that are an aircraft?

Correct me if I'm wrong, but adding extra cameras doesn't seem to solve this problem, each source reporting "almost everything moves" would make solving intersections and tracking them impossible, because each target candidate can be assigned to many changes pixels in consecutive frames. Unless some additional pattern detection is done, but again it's hard for very small objects.

[mapt]

> This is not much to determine what craft is approaching us.

Why do you have to do that exactly? Aircraft identification and aircraft detection are very different tasks. For detection, you need a tiny fractional difference in illumination (<1%) of one single pixel, that persists over time, and which shows up on two or three cameras separated for parallax.

The Youtube channel Consistently Inconsistent has been doing a series on optical detection, after an offhand Elon Musk comment.

https://www.youtube.com/watch?v=m-b51C82-UE

https://www.youtube.com/watch?v=zFiubdrJqqI

https://www.youtube.com/watch?v=YZkLQsv3huo

Anything detected can be subjected to closer inspection with radars, optical telescopes, and infrared telescopes.

[petschge]

You don't need to identify the A320 from 15 pixels. Once you see a 15 pixel signature that changes between subsequent frames you point the one extra camera on a movable mount with a 600mm lens on it at it. Now you get a few hundred pixels.

[LargoLasskhyfv]

Still making sounds. And usually not less than an A320. May be supersonic though. So even less time to adapt/react.

Anyway. Combine with microphone arrays, and your coverage is better.