The one feature point-and-shoots still have over DSLRs is the ability to shoot video clips. DSLRs are optimized in the other direction—the mirror setup, specialized auto-focus systems, etc. Hiroshi Terada's patent for a movie mode on DSLRs has one big innovation—a semi-transparent mirror—and then pulls a PS2-on-PS3 emulation trick, cramming in the components necessary to shoot solid video alongside the DSLR's regular components.
In order to shoot at 30FPS, Photography Bay notes that a DSLR's mirror would have jump up and down at least three times faster than Canon's EOS-1D Mark III. Terada's semi-transparent mirror doesn't move, but lets 70 percent of the light through to the image sensor, enough for video. The reflected 30 percent takes care of your continuous phase-detection autofocus.
The patent actually uses two different autofocus systems—one for still image shooting, and then another for clips, which has a wider AF threshold (DSLRs tend to have a very narrow one) and a slower, smoother AF speed (vs. the light speed you want for still images). A cropping function would take care of field of view changes that happen during AF shifts.
After Live View, this was the only place left to go—the question is when we'll see these first Zapruder-capable DSLRs, though the next generation doesn't sound like an unreasonable time frame. [USPTO via Photography and Gizmodo]
OK, but for markets like photojournalism that increasingly are shooting video alongside still, it seems the still frame grab capabilities of video cameras are sufficient quality for say, newspapers, and getting better, so I don't see the advantage of this besides marketing to amateurs moving up from P&S.
The idea of a semi-transparent mirror is, of course, quite old... it offers many more useful features than just allowing for HD video. ie. continuous autofocus without blackouts, less noise, less vibration, faster response, etc. It would be neat if canon were to reintroduce the pellicle mirror of the EOS RT and 1N RS in a 1D.. especially with the superb high iso these days to counteract the loss of light. I don't care too much for the video, but there are so many useful applications of using such a mirror.
floris wrote:
The idea of a semi-transparent mirror is, of course, quite old... it offers many more useful features than just allowing for HD video. ie. continuous autofocus without blackouts, less noise, less vibration, faster response, etc. It would be neat if canon were to reintroduce the pellicle mirror of the EOS RT and 1N RS in a 1D.. especially with the superb high iso these days to counteract the loss of light. I don't care too much for the video, but there are so many useful applications of using such a mirror.
And it also introduces the need to have a clean and damage free mirror.
Why not just forgo the mirror and use an extremely high resolution LCD panel as the viewfinder that pulls its image from the image sensor. AKA P+S camera design, from which "live view" has been borrowed from in the first place.
fizzy wrote:
...so I don't see the advantage of this besides marketing to amateurs moving up from P&S.
And they said the same thing about live view ;-)
I've played with LV on my 40D for a couple months off and on. Thought I might find some use for it. The camera LCD screen is just too tiny and dim to see what the hell is going on--the normal VF is so much better. Well, LV was somewhat useful tethered to my Mac but beyond the odd still life studio shoot, it's just another hardly used feature like the direct print button.
What would actually be pretty sweet is if they implemented one of those systems where it stores say the last 5 sec in a buffer, and then you press the button to save the buffer - that way you would never miss an exciting moment, without burning through too much memory. Of course, that would give you way more (still) frames than you could possibly want from a photographic point of view, but I'm sure it could be useful.
If I want video, I'll buy a video camera. I have a DSLR because I want to take stills. To me it's just more bells & whistles to add to the sales brochure list. Meh.
floris wrote:
The idea of a semi-transparent mirror is, of course, quite old... it offers many more useful features than just allowing for HD video. ie. continuous autofocus without blackouts, less noise, less vibration, faster response, etc. It would be neat if canon were to reintroduce the pellicle mirror of the EOS RT and 1N RS in a 1D.. especially with the superb high iso these days to counteract the loss of light. I don't care too much for the video, but there are so many useful applications of using such a mirror.
I've been bringing up the idea of a pellicle mirror (that is what Canon and Nikon have referred to it as) for years now for DSLRs. Since DSLRs don't have to advance film, there isn't any reason why they can't shoot at higher frame rates. It really puts a sock in those rangefinder people's aguments when they talk about not being able to see the image at the time of exposure, and I bring up pellicle mirrors. The idea of using it for high speed is actually very old, the Canon F1 and F1n high speed (10 and 14 FPS, respectively), and the Nikon F2 and F3 high speed (9 and 11 FPS I believe). Canon also made consumer pellicle cameras during the 70s as well, the Pellix and the Pellix QL. Very nice cameras.
dirb9 wrote:
Since DSLRs don't have to advance film, there isn't any reason why they can't shoot at higher frame rates.
Data throughput is a simple reason, if ignoring that, maybe heat, storage, and power could all limit the effective speed.
If you get a system that deal with all of those problems, you might have something that looks like the red one camera. It sometimes has issues with fast movement because it uses a rolling shutter (rather than a global shutter).
invalid2 wrote:
Data throughput is a simple reason, if ignoring that, maybe heat, storage, and power could all limit the effective speed.
If you get a system that deal with all of those problems, you might have something that looks like the red one camera. It sometimes has issues with fast movement because it uses a rolling shutter (rather than a global shutter).
1920x1080 full HD means 2 Mpixel per image. Now do 30 frames per second. The 1D(s) Mk II series could do about the same pixels-per-second throughput. So that's not the bottleneck.
Storage? Full HD will do up to 24 Mbps, compressed output at highest quality, which means just under 3 megabyte per second. With the biggest Pretec CF card of 48 GB (Samsung announced a CF card of 60 GB) you're looking at more than 4 hours of recording time.
... Now, moving to Quad HDTV (3840×2160 or 2160p at 8.3 Mpixel per frame) would be a challenge for dSLR and CF storage. For that we'd need HD storage.
dirb9 wrote:
Since DSLRs don't have to advance film, there isn't any reason why they can't shoot at higher frame rates.
invalid2 wrote:
Data throughput is a simple reason, if ignoring that, maybe heat, storage, and power could all limit the effective speed.
If you get a system that deal with all of those problems, you might have something that looks like the red one camera. It sometimes has issues with fast movement because it uses a rolling shutter (rather than a global shutter).
Tentacle wrote:
1920x1080 full HD means 2 Mpixel per image. Now do 30 frames per second. The 1D(s) Mk II series could do about the same pixels-per-second throughput. So that's not the bottleneck.
Storage? Full HD will do up to 24 Mbps, compressed output at highest quality, which means just under 3 megabyte per second. With the biggest Pretec CF card of 48 GB (Samsung announced a CF card of 60 GB) you're looking at more than 4 hours of recording time.
... Now, moving to Quad HDTV (3840×2160 or 2160p at 8.3 Mpixel per frame) would be a challenge for dSLR and CF storage. For that we'd need HD storage....Show more →
Sorry for not stating more clearly that I was listing limitations for a ~10MP, 30+fps camera - and expecting that the same limitations would hold for 21MP at similar frame rates.
If you want raw output (like with a dslr), you would need 10MP*12bpp*30fps=540MB/s (which would run to 1.8TB in an hour). Now if you can compress the data, you can get a lower bitrate, but then you need to make sure you can compress it fast enough. To pick a real world example, the "redcode raw" codec can run at 4k resolution up to 30fps with a compressed data rate of about 27.5MB/s (ref, CineForm ref). It is not able to handle faster frame rates at that resolution (I do not know why).
If you want to consider just HD resolutions, eg 1080p - please do look at SDI data rates (ref).
If you want raw output (like with a dslr), you would need 10MP*12bpp*30fps=540MB/s (which would run to 1.8TB in an hour). Now if you can compress the data, you can get a lower bitrate, but then you need to make sure you can compress it fast enough. To pick a real world example, the "redcode raw" codec can run at 4k resolution up to 30fps with a compressed data rate of about 27.5MB/s (ref, CineForm ref). It is not able to handle faster frame rates at that resolution (I do not know why).
[...]
Ah, but it is. The RedOne will do 2540p at 30 fps RAW if you record to external RAID. The package of two laptop HDs in RAID 0 isn't fast enough to take more than 4K with lossy codec, but that doesn't mean it can't be done.
The company decided to discontinue the option for RAW 2540p output for now because it was technically difficult and the external HD array solution made it cumbersome, but that doesn't mean it couldn't be done in the first place.
If we assume lossless RAW compression of about 1 Mpixel to 1.2 Mbyte (seems fair, Canon RAW is comparable) then 30 fps at 12 Mpixel will mean around 400 to 450 MB/sec output. Group 4 high-end hard disks together in RAID 0 and you'll easily push this much data onto hard disk. Two 3 Gbit SAS links should do the trick.
The idea of using it for high speed is actually very old, the Canon F1 and F1n high speed (10 and 14 FPS, respectively), and the Nikon F2 and F3 high speed (9 and 11 FPS I believe). Canon also made consumer pellicle cameras during the 70s as well, the Pellix and the Pellix QL. Very nice cameras.