3 channel 36 bit 1280 X 720 low $ camera - Viper?

[Steve Ipp]

History_____________________________________________

Dear all, I have been reading this forum for more than a year by now in search for an answer: how to assemble a low cost high bit high resolution camera.
I also sent dozens of e-mails to cmos/ccd/camera manufacturers. I spent 8 months doing this.
By now, I am upset at me, waisting valuable time, I am upset at JVC for waisting my time comparing it's HD1, I am upset at Panasonic telling me that the $3000 flash camera will be available in year 2006.


Good results :)______________________________________
There is a cmpany out there that sells monochrome 1.3 megapixel (1280x1024) CMOS cameras with USB2.0 interface.
I quired whether they can help me out with my project (3 CMOSs/prism/final RGB stream). They replied very nicely, I am pasting the whole letter below.
(
P.S. for those people, who don't get a clue what I am up to, here's a lead:
1. Take three monochrome sensors, glue them to a prism (RGB)
2. Capture each monochrome stream on a powerful computer (2.8 GHz should be enough)
3. Premultiply each captured stream (12 bit X 3) with R, G, B
4. Get yourself a candy - 36 bit 1280 X 720 24 fps
)

Here is the reply:

Dear Steve,

Thank you for your interest in SMX 150 cameras. Please let us know what
kind of images you would like us to prepare for you, e.g. what
illumination, what kind of lens, speed, magnification, etc.
The price for SMX-150 M is $1050 and for SMX-150C is $1100, which
include tripod adapter, application software, DirectShow and Twain
drivers, and SDK software.
For two cameras with a PC with 2.8 GHz we have not seen any
deterioration in speed, both camera can work at 33 fps for 1000 by 1000
images. For three cameras you may need an even faster computer. If
needed we provide you with modification in the driver software so the
three cameras would produce a RGB stream.
If you need help with special optics for combining thee monochrome
cameras into a Color RGB camera, please let us know. For volumes of 10
or more we would be interested to produce 3 sensors RGB cameras.

Best regards
Farhad Towfiq, Ph.D.
_________________________________________________

The last lines made me post this stuff here.

Result:________________________________________
Regretfully, I didn't order the 10 cameras the guy wanted me to. The reason is simple: I don't have a spare 45 to 60 K. But asking nicely again would do the trick.
Here's the deal: if you are interested, post here, I will continue the dialog.
Please, don't mail them yourselves: they might freak out if the interest in RGB exceeds their expectations.
If that happens, we will have the same 20 000 HD solution SONY is pitching. Let it happen for 4000$ or so.
here they are
nntp://vvv.sumix.com
Having at least 10 people on board, I will promise the guy all 10 cameras will be ordered in a matter of a week.
I haven't asked for the price of the final camera, if you turn out to be interested, I'll check on it and post here.
Let me know what requirements you have other that I described above.

Thank you for reading.
Much respect,
Steve

[James Ball]

I am interested.

I'd like a ton more detail. Please email me. like you, i have been on the same fruitless quest and would be glad to share what I have so far.

The main thing I'd like to see is a bayonet mount so that we can use off the shelf interchangable video lenses.

[John Cabrera]

Wait, though... why wouldn't someone not purchase one of those single chip color versions. I mean I know it's not gonna be a 3 channel RGB image, but functionally quality wise it should blow out of the water just about any single chip camera out there, including the JVC MiniHD system. High resolution raw 24bit color image streams at almost any frame rate direct to disk? For about a grand? Damn... sounds good to me... especially with an Aldu35 setup.

Am I missing something here?

John

[Luis Caffesse]

"There is a cmpany out there that sells monochrome 1.3 megapixel (1280x1024) CMOS cameras"

If I'm not mistaken John, what is being referred to here is taking 3 of these MONOCHROME sensors, and splitting the light with a prism into Red Green and Blue (in the same way most 3 chip cameras work now).

If you were to make a 1 chip camera with these sensors, there would be no way to recreate the color information. With a prism, the light is divided by color before reaching the chips, so although the chips may be monochrome, it is possible to recreate the color by knowing which chip is collecting the Red light, Green light, and blue light.

Then again, I may be way off. Most of this stuff is over my head.
:)

Sounds interesting though.
Keep us updated!

-Luis

[John Cabrera]

I know exactly what he's trying to do. He wants to make a three chip camera out of three of their one chip mono cameras. But in addition to their mono camera, they also sell a single chip color camera, and judging from the sample images it shoots pretty nice color at that. I realize that if he can get enough people together to convince them to build a 3 chip version of one of these things it would be fantastic, but I was pointing out the fact that they already sell a one chip color version that seems to greatly rival any other one chip camera out there.... in price, quality, and functionality.

Also, I know that many of the chips being made now for higher end single chip cameras are actually three layers that act exactly like three chip cameras. Are their color chips three layer chips, I wonder.

J

[Steve Ipp]

John,

Thank you for comments, you're right - some manufacturers are starting to come up with new designs incorporating one CMOS sensors comprised of 3 layers.
This technology is revolutionary but Foveon (the patent holder, if someone doesn't know) is very paranoid about the licensing aspect of the invention. There are only 2 still cameras on the market (one of them hasn't been put on sale yet) that have the chip.
As for the one chip color industrial camera from the manufacturer I mentioned in the post, - you're 100% on point; 1000 $ USB cam puts all DV crap in shade (although most DV cameras outperform in low light).

Luis,

Thanks for posting, this monochrome stuff is working. Just for fun, try taking a three channel high res image into Photoshop. Separate it's RGB into different images. Take a note which channel is which. Convert them to monochrome. Multiply each of them by the corresponding color (R, G, or B). Paste back to the channels.
You should have something you started with.
There was one guy on another site who tried this trick with GR_HD1 (he used only one monochrome image to create three).
He said it works, but takes too much time.

[Jef Bryant]

Be careful. I researched a number of similiar machine-vision-style cameras that seemed promising even after carefully reading the specs and communicating with the companies.

I discovered that it's difficult to find one of these cameras that can shoot film-style, meaning a 48th of a second exposure at 24 frames per second (I'm assuming that's what you want). Many state that they can be programmed to shoot 24fps, and that you can program the exposure to be 48th of a sec (around 21 ms) or even longer, HOWEVER, when you set the exposure times to around 21 ms, the readout time of the chip also increases, and the max frame rate is forced to decrease. So you can shoot at 24fps, and at 48th of a sec exposure, just not at the same time.

The last camera I checked on had a chart of fps rates and image sizes, but the chart was made with a 3 ms exposure time setting.

So before you spend any money, be sure the camera can do both of these things simultaneously. It can be difficult to wrangle this kind of specific information from the sales reps.

Also, you will want to use the global shutter, not the rolling shutter which exposes the chip in a totally different way, but you can only use the global shutter when you are applying an external trigger (if I'm reading the manual correctly). So you'll have to rig up something to trigger the camera 24 times per second. They offer the trigger cable, but I think you have to come up with your own triggering unit.

Maybe I'm telling you things you already know.

If the camera actually meets the 48th of a sec at 24fps criteria, I might get one myself, but I'm only interested in the single chipper.

[Steve Ipp]

Thank you Jef,
I will definitely bring up this issue before the engineers. This is a very valuable piece of information I have omitted in describing the parameters of the future camera.
In this regard, could you expand on how to check the quality of the output?
Should I ask the company for several hundred frames output in 24 fps mode or is there another way?
I hope you will find time to look at how we're doing here every once in a while. The final solution could make a lot of compositors and smal SFX studios happy.
Thank you again,
Steve

[Robert Knecht Schmidt]

You might be able to do the triggering from the serial port of the capture laptop, or, depending on the requirements, a little TTL logic may be sufficient.

[Steve Ipp]

As the trigger point has just been cleared, I will return to shutter.
Which do you think is more favourable - software or mechanical?

[Robert Knecht Schmidt]

"Which do you think is more favourable - software or mechanical?"

I assume by "software" you mean "electronic." I'm not sure your question makes sense in this context.

Regardless of whether or not a mechanical shutter is used, the camera will need to be electronically shuttered in order to gather up the intensities collected by all the pixels (readout phase). I'm not sure what the advantage would be of having an additional mechanical shutter unless some specific effect is desired (e.g. the infamous Saving Private Ryan look).

Mechanical shutters are normally only used in motion film cameras and still cameras. They're necessary in motion film cameras only to prevent a blurring exposure during the time that the current frame is shifted away and the next frame is clawed into place. Some video cameras implement mechnical shutters to similar effect to prevent smear during the readout phase, but in any case, the utility of building your own mechanical shutter system and trying to synchronize it with the electronic shutter of the off-the-shelf sensor apparatus is going to depend on the technology used by the sensor apparatus. If your vendor is still willing to provide you sample images prior to purchase, ask for the camera to be focused on a bright light source (e.g. car headlights at night) to evaluate smear.

The "rolling shutter" (ERS) listed on the spec sheet of the camera in question is a feature common among CMOS sensors and is an electronic process--it doesn't mean that there's a mechanical shutter rolling around a la a motion film camera.

[Steve Ipp]

Thanks for tolerance, Robert
Seems like prior to getting a proper hardware system together one needs to have more experience not only with electronics and rapidly acquired knowledge of digital image math but also solid understanding of working film cameras. Getting knowledgeable people together was the point of bringing the whole issue up.
Yes the vendor is still interested, they want me to comment on the following topics:

1. Portability
2. Inferior CMOS sensitivity (compared to CCD sensors)
3. Connectivity with capturing system (three options - Camera Link, IEEE 1394A or 1GB Ethernet)
4. Data rate (no compression, they estimate it to be around 82 MB/s)
5. Light splitting prism (not cheap)

Apart from that, the guys need at least 3 month for development.
Another thing is that the monochrome CMOS sensors output 10 bit, not 12.

Here's what I think (please, do correct me, if I lack knowledge in the following answers).

1. The camera should be portable. For that reason the only option is 1GB Ethernet wireless.
2. Because each pixel on a CMOS sensor has several transistors located next to it, the light sensitivity of a CMOS chip tends to be lower. Many of the photons hitting the chip hit the transistors instead of the photodiode. Thus, compared to a CCD chip, the same amount of pixels on CMOS would count to something like 100 to 70. The less lux you have, the more smear you'll have in the picture. Having this ratio around only tests will answer the question of image usability in low light conditions.
3. Refer to No 1.
4. To capture the stream you need a powerful system. I tested speed on my dual Xeon 2.4 HGz with onboard SCSI and software RAID. The tests show something like 54 MB/s. This problem fades away once you install a PCI SCSI card with hardware RAID level 0 (if not mistaken) on it. I have a Syborg 2.0 (an extinct PC HD compositing software) manual on system assembly. I used it as a blueprint for building up my workstation. The manual confirms on 100 more MB/s transfer rates with the hardware RAID and dual Xeons.
5. This could be a pain in the ass, if the company doesn't have OEM connections with optics manufacturers, it could turn out to be very costly (5 K US or more).

Drawing the bottom line, the whole system wouldn't be cheap.
Here are my estimates:

1. 300 US - Gigabit ethernet hardware ( I am in Taiwan)
2. 3 - 6 K (this is more than aproximate, - it could be more than that - the prism and development costs will be the major factors; I still need to reply the vendor and ask them for an estimate cost) camera assembly.
3. Lens (you count, - it might be anything depending on what you fix to the bayonet mount. I am considering using some good glass from second hand digital still cameras with homemade aldus).
4. Ground glass assembly (do it yourself style or 8000 US final solution)

Not so cheap after all.
I will get the camera aproximate cost after working all this stuff with the technical folks from the sumix. All will be posted here.
Untill that I do hope for your comments on the semiutopia. There might be ways to cut the corners (e.g - going for one chipper considering that the folks from sumix are the only ones I spotted who actually do care to push the development of customer inspired product).

Much respect,
Steve

[Adam Burtle]

how large are the chips? if someone would start making 35mm HD chips, then people could shoot 4:4:4 HD and get all the other characteristics of film (dof, angle of view, etc) and then color correct to their own tastes. i'm sure these are probably 1/2" or smaller though :(

[Steve Ipp]

Yes Adam, for price considerations the manufacturer proposes 1/2" size of the chips. The development hasn't started yet, we're figuring out the technical characteristics. There is an option to go for larger/more expensive 2/3" chips.

[Steve Ipp]

here's a letter from sumix:

.... About three sensor camera we did some thinking. Seems retail price of
$2700-$3000 is feasible.
There are several issues. One issue is the sensor. If external trigger
is not required we should use a different sensor. The sensor size is
also critical. The Fillfactory IBIS5A is a 2/3" sensor. We can use
smaller sensor like a 1/2" micron sensor. This can reduce the price of
lenses as well as the prices of the components and also overall size
will be smaller. Both sensors have 10 bit A/D. 12 bit can only be
obtained by multiple exposure and sampling which drops the frame rate.
The only sensor that has 12 bit is Foveon which is not acceptable for
its slow frame rate.
The other issue is some form of compression to reduce demand on disk
storage. Perhaps you accept only lossless compression.
Also, we need to decide what kind of lens interface is the best. In
industrial vision we use C-mount objectives which are economical for the
sizes of sensors we use.