Slashdot Mirror
Canon Shows the Most Sensitive Camera Sensor In the World
An anonymous reader writes "Canon announced today that it successfully developed a super high-sensitivity full-frame CMOS sensor developed exclusively for video recording. The new Full HD sensor can capture light no other comparable sensor can see and it uses pixels 7.5 larger than the best commercial professional cameras in existence today."
There doesn't seem to be a gallery of images, but the video demo (direct link to an mpeg4) makes it seem pretty sensitive.
That's the fundamental rule of how cameras function. Without such a limitation, you'd need incremental exposure timing/capture which I don't believe any sensor's can perform, then you need to actually process the HDR'ness of the image, which is quite frankly very subjective. One may choose to blind the viewer with the light shining through the window, or one may want to see the house across the street. This is an artisitic quality that needs to be supported regardless of which technology you choose. In the down to earth point of view, you may look into bracketing, which can at least support HDR from most decent SLR's, but of course even those techniques require two shots, meaning basically absoltely still shots. The real HDR shots are taken with prism splitters into two bodies, but that means two identical cameras with a custom expensive setup... Well, nobody said the perfection was cheap.
Bye!
Err, of course the sky is blue under moonlight: it's just reflected sunlight, after all (but see below).
The problem is that the Moon is much fainter than the Sun and thus the overall light level is low. So low that it doesn't significantly activate the colour-sensitive cones in the human eye, meaning that you only really see with the rods in black-and-white.
But take a long exposure with a camera (or a video frame rate with this Canon sensor), and the blue will most definitely come through.
(Actually, the moonlight-illuminated sky is slightly bluer than a sunlight-illuminated one, as the Moon's slightly brown-ish colour first imprints its spectral dependence on the sunlight which bounces off it. That light is then Rayleigh-scattered off the molecules in the Earth's atmosphere, imprinting the well-known 1/lambda^4 dependence which makes the sky blue).
The newly developed CMOS sensor features pixels measuring 19 microns square in size, which is more than 7.5-times the surface area of the pixels on the CMOS sensor incorporated in Canon's top-of-the-line EOS-1D X and other digital SLR cameras.
I guess this is to collect more photons in low light conditions. Of course this means that sensor is physically larger, but that's not a problem for Canon, they have made medium format cameras in the past.
Have you ever taken a shoot with a 50mm f:1.5 on one of today's pro cameras? If you multiple the light available by 56 times (increasing the pixle size by 7.5) you are looking at a shit load of light.
I can shoot nice outdoor night pics under a full moon with just an F5 @ 5 seconds, drop that down to a F1.5 and that's more than 8 times the light, add this new sensor and that's 280 times more light! or about 6 FPS and note this is on a canon 5D MK I that's almost 7 and 1.2 years old..
Add in the current 3 and half year old generations improvement on the ISO and that goes from the 1600 I shoot at, to 6400, or 4 times the light and you get 24 FPS..
So thats with jsut the sensor.. If they where using F1.2 or even F1 lens and one would expect when showing such a beast and 30 FPS seems like no issue at all..
You have 5 Moderator Points!
Which Helpless Linux zealot/MS basher do you want to mod down today?
That need not even be the case. You could still do it in 35mm, 1080p HD video is roughy 2.1 megapixels, where as the EOS1DX is 18.1mp.
So there is definitely enough room to make pixels 7 times larger than a EOS1DX
Sometimes I wish I was a plumber, then I'd know how to deal with other people's shit.
It's a 35mm full-frame sensor.
It's also explicitly intended "exclusively for video recording" and mentions "full HD". Which would mean - assuming I'm reading between the lines correctly - that the resolution is 1920x1080 - ie. 2 megapixels.
Of course this means that sensor is physically larger
The sensor isn't physically larger. The specs say it's a full-frame 35mm sensor, and the photo of the prototype camera shows it with a standard EF lens mounted: a larger sensor would need medium or large format lenses, and it would be pretty much dead on arrival in the market if you had to go out and start buying medium or (God forbid) large format lenses to feed the thing. Half of the allure of Canon for video, after all, is that you can reuse your still EF lenses, and demanding huge format glass for HD video would be absurd.
The reason the photo sites are so much bigger in this sensor, presumably, is because the resolution is much lower than Canon's still SLR cameras. It doesn't give the resolution, but since it was only described as capturing "HD video," I wouldn't be surprised to find that the sensor's native resolution is that of 1080p video: 1920x1080 pixels, or about 2 megapixels. The 1Dx, on the other hand, has a native resolution of 18 megapixels.
So far, Canon (and more recently Nikon), have been allowing users to record HD video on their SLR cameras by scaling their massive native resolutions way down to a size that you can reasonably encode and cram onto a memory card in an SLR form factor. This approach, on the other hand, seems to be to build a sensor with a lower native resolution suitable for HD video at the same size as the larger SLR sensors, so you don't have to do any down-scaling and you get massive photo sites, which gives you a huge advantage in sensitivity.
Sensitivity and dynamic range are separate things. You can have an extremely sensitive sensor, pushing an equiv of ISO 12800 or even more, but dynamic range may only be 8-10EV
Even the best of the best have around 13EV of dynamic range(eg Nikon D7000) at ISO 100. As you increase ISO, the dynamic range suffers, and noise increases. Getting to above 14-15EV is very very difficult. You can do it in post processing(HDR combination of multiple exposures)
My Aurora : http://www.youtube.com/watch?v=o91ZsGwJYyg
FB : https://www.facebook.com/TanveersPhotography
Yes, but then you would need special lenses. This sensor works with 35mm (full-frame) lenses.
"A week in the lab saves an hour in the library"
If dynamic range is important to you, you may be interested in Rambus' new technology:
http://www.dpreview.com/news/2013/02/27/rambus-shows-binaryt-pixel-sensor-technology-for-expanded-dynamic-range
Could not doing 3x3 binning on the existing 18MP sensor (if the controller supported it) produce similar results?
19-micron pixels seem big if you're comparing them to DSLRs, where everything has to fit into a nice little portable package. But it's not at all an unusual size in science-grade detectors used for astronomical instrumentation. At work our instruments use detectors with pixel sizes ranging from 13.5 to 50 microns.
I might be a little more impressed that they're doing this at video frame-rates, and without cryogenics...
Village idiot in some extremely smart villages.
Well, the credits of Barry Lyndon include a special thanks to Zeiss for the lenses.
Just goes to show, it's all in the glass - you can have as many megapixels as you want in the sensor, but if your lens isn't up to it, you're throwing away the potential of all those pixels.
They sentenced me to twenty years of boredom
Especially with the sky being blue from the full moon alone.
is the same as the AC who then posted this one:
I obviously didn't object to it being that hue, dumbass. I objected to it being *that* bright. It was a day shot. And obviously so.
then I'd say that's exactly what you did say.
And as for your assumption that I'm an American ... well, you haven't got a clue, mate. You're many thousands of kilometres off. There are other countries in the world where English is the native language, after all. "We have 2013" indeed.
Sometimes I really do wonder whether /. is worth the trouble.
That's the comparison you saw in the video. Binning is generally how the down-sampling is done.
Moonlight is just reflected sunlight, you just need more of it to make the colours come out.
E.g. see http://www.flickr.com/photos/dansdata/3074862610/ for an example - this photo was taken under a full moon, 30 second exposure.
By "larger width" lens are you meaning larger aperture, smaller focal length or larger image circle?
All of these things have effects on the image and have practical limitations. (for example change the image circle area to be larger and you need a new format, all the old 35mm lenses they have can't be used for the larger frame unless you like black edges).
And you can have the best glass in the world, but if your sensor is from say a kodak dcs 620, that is just a complete waste. The whole system has to be suited, one weak link kills quality.
They use another trick to take care of this, though: "microlenses", miniature optics in front of each pixel that channel light away from the insensitive regions (the data paths) and onto the actual light-sensitive pixels. A recent advance is "gapless microlenses", where nearly all of the light incident on the sensor winds up falling on some pixel or other.
After size, the big advantage of using 35mm is simply being able to use the overwhelming piles of 35mm lenses out there. If you have some big fancy camera you're going to need big fancy lenses. But if you can put the big fancy sensor in a normal camera, then you've got a whole new market there; people who can afford or already have 35mm lenses but can't afford all the kit for a bigger, more expensive camera. That's who this sensor is for. I can't help but think that it would be an excellent thing to sell to schools at only a nominal profit. Then students could buy an SLR-format camera with the same sensor and accepting the same lenses as what they learned on, when they graduated. Get a sort of Apple-like function going on.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Learn some electronics. Signal increases directly with sensor area, noise with the square root of area (generally).
Contribute to civilization: ari.aynrand.org/donate