36
$\begingroup$

When I first saw Google Maps satellite view many years ago, I remember being greatly impressed (and somewhat surprised) with the image quality, considering the photographs came from up there in space.

I later learned that only some of the texture is from satellite imagery, and the higher resolution content is usually sourced from aircraft flyover in metropolitan areas.

Now I'm curious about just how good the state of the art in satellite imagery is today. For example, if a person of interest was looking up into the sky on a clear day, would the photography be good enough for facial recognition? How many centimetres on the ground per pixel in the image at maximum zoom? Is this technology still improving, or is there some sort of physical limit which now prevents increases in resolution?

|improve this question
$\endgroup$
  • 1
    $\begingroup$ Just don't look up. $\endgroup$ – Mark Adler Mar 2 '14 at 20:35
  • 7
    $\begingroup$ Obligatory: "I could tell you but then I'd have to kill you" $\endgroup$ – Carl Witthoft Mar 2 '14 at 22:08
  • 1
    $\begingroup$ @AnthonyX: Optical mirrors are quite heavy, so a boom that can keep them located accurately enough is quite hard. You are correct that the effective f/stop is very slow for virtual mirrors of this type, so the exposures would have to be fairly long. Then people move.... $\endgroup$ – Ross Millikan Mar 3 '14 at 3:49
  • 7
    $\begingroup$ Doesn't exactly answer your question, but related (and a great read): XKCD What if?: If the Hubble telescope were aimed at the Earth, how detailed would the images be? $\endgroup$ – IQAndreas Mar 3 '14 at 6:17
  • 1
    $\begingroup$ @CarlWitthoft ...from orbit. $\endgroup$ – SQB Mar 3 '14 at 7:07
30
$\begingroup$

There are limits. For one, there's atmospheric effects that scatter light in visible wavelength spectrum. You might be able to penetrate clouds and haze easier in the lower end of the spectrum and towards the infrared wavelengths, and those might still be usable for facial recognition though. Another limit is aperture of optical equipment used to take photographs, in case of telescopes, that's given by the Dawes' limit and is derived from Rayleigh criterion to compute angular resolution of imaging devices. So the end resolution depends on weather, imaging wavelength, atmospheric scattering, distance, and diameter of primary mirror (aperture). Build quality obviously also has a major role.

But let's first see what's the absolute minimum resolution we'd require to recognize facial features and successfully identify individuals. According to this article, to do that, we'd need to resolve a human face to a resolution of at least 40x40 pixels. So if we say that an average adult human face is 20 cm wide, we should resolve to within 5 mm. At also extremely low LEO orbits that might be used by some surveying satellites, say 200 km above the surface of the Earth during perigee of highly elliptical Molniya orbits, this gives us a required angular resolution of $3.0\cdot 10^{-8}\text{ rad}$, or $0.00618794419\text { arc}$  (in radians and arcseconds, respectively).

Using Dawes' limit, we can then calculate theoretical minimum telescope's aperture (in centimeters):

$$D = \frac{11.6}{R} = \frac{11.6}{0.00618794419} = 1,874.6\text{ cm}$$

Or roughly 18.7 m (738 inch, 61.5 ft, 20.5 yard) diameter telescope. Not impossible, but such huge telescope would certainly be visible even with a naked eye (remember, that's minimum diameter, it would be by far taller than that) if it caught reflection, and that's usually not an option with military spy satellites we might not be aware of, while there certainly aren't any such commercial and/or scientific telescopes in orbit.

Considering I took the most extreme and ideal case examples into consideration for the sake of argument, and the numbers are an order of magnitude smaller than what David Hammen came out with in his answer because of that, the answer is still:

No, there is no such optical telescope in Earth's orbit with required resolution to identify individuals by their facial features. If it was, we'd know about it.

One thing that shouldn't be forgotten is that satellites can't simply change orbits and be at any place and at any time they'd need to be, to track and identify individuals even if they could do that by their facial features alone. If someone is going to put so capable optical equipment in the skies, it would more likely be on inconspicuously painted blimps / aerostats that could silently hover at a lot smaller altitudes, image targets from a shallow angle, and follow their movement easier. And all of that with a lot cheaper, smaller and harder to notice equipment. Or use helicopters, drones, CCTV,... or good old boots on the ground.

|improve this answer
$\endgroup$
  • 3
    $\begingroup$ +1 for mentionning weather! Even if we had some gigantic optics in orbit, atmospheric turbulence and scattering would still be the limiting factor! $\endgroup$ – PhilMacKay Mar 3 '14 at 15:23
  • 1
    $\begingroup$ You are leaving out aperture synthesis, though. $\endgroup$ – kert Sep 12 '17 at 2:42
27
$\begingroup$

would the photography be good enough for facial recognition?

Not yet. It's not even close. Facial recognition requires 50 to 100 pixels between the eyes, or on the order of 1 millimeter resolution. To see that kind of detail from a distance of 250 kilometers using blue-green light (500 nm) would require a lens or mirror that is 125 meters in diameter. Note: I'm being very generous with that distance of 250 km and with the use of blue-green light. For the physics involved, see http://en.wikipedia.org/wiki/Airy_disk.

The KH-12 satellites supposedly have a resolution of 6 inches or so. Given the sampling theorem, that means a one foot resolution in practice. You can't even see that it's a face that you're looking at with that kind of resolution, let alone whose face it is.

is there some sort of physical limit which now prevents increases in resolution?

Yes. That light is a wave phenomenon places limits on the resolution of imaging equipment. High resolution requires big lenses or mirrors. See the article to which I linked early.

|improve this answer
$\endgroup$
  • 1
    $\begingroup$ @DanDascalescu - The OP is concerned with satellite imagery, not aerial imagery in general. Regarding reading license plates from a satellite: That's a myth. The KH-11B satellites reportedly have an optical resolution of 5-6 inches at the subsatellite point. That is insufficient resolution to read a license plate mounted horizontally on the roof of a car, let alone a real license plate that is mounted vertically. A satellite would have to be looking at an angle rather than to the nadir to see a license plate. That increases to the object and increases atmospheric problems. $\endgroup$ – David Hammen Mar 3 '14 at 13:32
  • 1
    $\begingroup$ Besides what the related xkcd says (and we don't know how much improved actual spy sattelites are) I just want to mention that when googling for something like "low pixel face recognition" the results suggest that in good conditions its possible to get some valuable information from images of only 25x25 pixels of the whole face. Adding some machine learning/tracking algorithms on top of this makes lots of recognition/tracking much less theoretical. Of course, only if you run around, looking up all the time... $\endgroup$ – PlasmaHH Mar 3 '14 at 15:01
  • 1
    $\begingroup$ I wonder if superresolution is being or could be used to get around some of the physics limitations? $\endgroup$ – M. Dudley Mar 3 '14 at 20:34
  • 1
    $\begingroup$ @PlasmaHH 25x25 is only good to recognize it as a face, say in point and shoot cameras, and even those err. 40x40 that I used in my answer is barely enough to distinguish between facial features of know individuals, so arguably good enough to determine that 1) yes, it's a face and 2) in a group of known people, it's x likely to be of a person y, where x is good enough. Any real facial recognition, such as for proper biometrics, would require far greater resolution than that. Having worked with image processing, 50 to 100 pixels between the eyes seems about bare minimum to me. $\endgroup$ – TildalWave Mar 16 '14 at 23:26
  • 1
    $\begingroup$ @M.Dudley Super resolution techniques are constantly used with satellite imaging and astronomy, essentially with any long focal length and/or exposure imaging nowadays, even live during some sports events broadcasts, in quad HD. Heck, even your smartphone camera likely has some resolution enhancement, noise reduction, auto-exposure, and other IQ algorithms built into its controller chip. For the answers here, that changes nothing, it's already accounted for by assuming unlikely yet optimal conditions. $\endgroup$ – TildalWave Mar 16 '14 at 23:39
2
$\begingroup$

This is of course theory-craft since I'm sure if satellites do exist that can resolve your face, read a license plate, or make out the time on your watch or a phone number on your iphone, we (the public) would be the last to hear about it.

I believe its worth noting that you can achieve the same with lower diameter lens simply because of how fast the satellite is moving or by quantity or both.

Explanation:

2 x 13 meter telescopes taking a photo of the same object at the same time equates to the same as 1 x 18 meter telescope. 4 x 9 meter telescopes can do the same as 1 x 18 meter telescopes. I am sure there are diminishing returns, but you get the point...

However

1 satellite at 1/2 meter lens moving at 25,400 feet per second can take 160 unique (different location) photos in 1/100th of a second. Since atmosphere, heat, and other contributors to atmospheric distortion do not change (enough) in 1/100th of a second, it's not really any different than taking 1 photo with a much larger diameter telescope.

One may argue the technology "doesn't exist to do it" but the truth is it does. Consumer technology available (like the camera they use on slow mo guys) is a high resolution 10,000 fps camera. I'm gonna go out on a limb and say technology available to the Defense Departments / NSA / Etc are better.

Another thing to note is the same technology used for atmospheric correction (the laser that draws a false star) can be used to take satellite photos. The satellite simply needs a landmark, or something to which the shape is known to be able to adjust and remove atmospheric effects.

I'm sure it'd be expensive.

The NRO satellites can coast as much as $9 billion though.

|improve this answer
$\endgroup$
1
$\begingroup$

The limitation of 10 cm or so here only applies for a single mirror. I don't know if it is possible but if you can fly two spy satellites in formation and combine the images using optical interferometry and given that the images from Earth are very bright- maybe, just maybe, they could improve on that?

So, have to say, I don't know, rather than that it's totally impossible. This is not my own idea though obvious once you think of it, I got it from the conversation on Snopes here.

|improve this answer
$\endgroup$
-1
$\begingroup$

Considering 2008 satellite imaging can resolve surface features down to 0.41 meters in size, I think you can be pretty sure that by now (2014) if you are sitting in your backyard reading a newspaper, the govt can know which paper you are reading.

|improve this answer
$\endgroup$
  • 3
    $\begingroup$ Welcome to SpaceEx.SE! A hyper-link to the source for your note (2008 resolution down to 0.41m) would be good to have. Having said that, it's not just the resolution - there's also the little bit unless a satellite is at GEO, it is skittering along at several km/s. $\endgroup$ – Everyone Mar 3 '14 at 9:10
  • 4
    $\begingroup$ 0.41 meters -- that's GeoEye-1. Link: en.wikipedia.org/wiki/GeoEye-1. Reconnaissance satellites do better than that, with about one centimeter resolution. It's a mistake to think that all technology advances at the rate at which computers are improving. Space technology, for example, doesn't. $\endgroup$ – David Hammen Mar 3 '14 at 13:41
  • 1
    $\begingroup$ That old comment of mine is incorrect. Reconnaissance satellites are believed to have ten centimeter resolution. The limitations are physics-based, and physics (unlike computers) hasn't changed. $\endgroup$ – David Hammen Sep 8 '17 at 16:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.