New DIY HD Cinema Camera Project

[Bob Tasa]

Did it end move or lost interest?
Just curious as I was waiting for some more info on
any progress.

Since 5 months ago have there been any new chips or revelations?

Bob

[Jose A. Garcia]

I guess Sumix offered what most of us were looking for and as I've got a lot to shoot and little time to do it, I'll just start by buying the 12a2c cam. After that I'll think about finishing this project.

[Alex Stewart]

Actually, Im still working away. Ive got my head board working on a logic analyzer implemented inside of the FPGA dev board Im using. This means I can see pixel data in byte form coming from the imager at 48MHz. The current challenge is storing this data stream, so I've been working with a SDRAM (8MB) to buffer a frame so I can attempt to offload it to a PC and get a "frame".

Full time student status makes progress on this task a little slow ;-)

One thing Im also planning is to use experiment with a cypress part (FX2) to stream raw data across USB 2.0.

So is anyone else still working on anything?

Also out of curiosity mainly, would anyone out there be interested in a kit of my final revision of the project?

My current laundry list of pieces for the kit/ final revsion are:
- Blackfin BF537 DSP
- Cyclone II FPGA
- 64MB SDRAM
- some amount of FLASH for firmware...
- 100BaseT ethernet
- MT9T001 Micron Imager

I figure between the blackfin and the DSP blocks in the cyclone there would be more than enough head room to run some good compression schemes to dump the video across ethernet (or USB 2.0)

Im rambling... by for now!

[Klas Persson]

I've been thinking of a great, at least to my knowledge, way to do a digital cinema camera. If you take a magazine from a 16mm(or 35mm for that matter) camera that has the sprocket mechanism in the magazine(Like the NPR) and put all your digital cinema diy products in there you'd esentially have a complete camera for very little money and work.

All the "camera" stuff would already be done in the body. You could just snap on your digital magazine to your body and shoot digitally or your ordinary magazine for film.

Lensmount, shutter, optical viewfinder, you name it in camera. The magazine just holds the sensor, computer, recording media, battery and connectors(Maybe a small lcd on the side for info). You could even have a switch that is switched by the film feeder sprocket so that the computer knows which framerate the camera is running at. Some kind of weird fusion of computer and mechanical age.

Now I do not know how a cmos sensor behaves but imagine that you take information from it at a regular rate, say 24/s, and you have a mechanical rolling shutter in front of it?! If so it seems it would work perfectly with a software that takes information from the sensor every time the switch I wrote about above registers a new frame.

It seems unnesecary to reinvent camera functions for a digital cinema camera when you more easily could just change the media.

You'd esentially just attach your s-16 sized cmos on the pressure plate inside the magazine. (Maybe with some sort of protection that keeps the pressure plate with the cmos inside the magazine until you snap it in place so that you wont scratch the sensor at attachment. Don't know exactly how that would be done but seems doable.)

If you don't have that much money you could just buy a r16 camera and shoot cropped until you can afford to turn it into s16.

Now I might be missing some major problems here as I'm not a sensor expert by any means but to me it seems like the perfect solution.

[Jose A. Garcia]

Alex, may I ask if you're already controlling the sensor with the FPGA, why do you need a computer? The ideal solution would be to have an independent camera.

[John Wyatt]

Klas -- I remember several years ago there was an attempt to convert 35mm SLR still cameras to digital by placing a panel-type unit (which had a sensor in it) into the standard 35mm pressure plate of the camera. I think the product changed names at least once but was called something like "efilm" or "SiliconFilm". The product didn't really take off when DSLRs became more affordable. I wonder if we could use an idea like that: just take an SLR still camera body and find a sensor which could be mounted in there, cabled out for dumping the data to a laptop or small computer? Like a glorifed webcam. I know some of these left field ideas are more problematic than just getting an existing video camera! But I think the spirit of these Alt Imaging threads is to ask questions and see if there's equipment and components out there, probably designed for other purposes, that might be used in some way to create DIY cameras (at a low cost). Sometimes I think a camera producing distinctive images can be more interesting than commercial cameras that produce technically better images, though obviously it depends on the type of projects you have in mind for them.

[Jose A. Garcia]

I have 2 old canon SLR cameras ready... If we place the sensor in the right position we'd also have autofocus if needed.

The only problem I see is that this is just a case. We'd still need either an APS-C or a full35mm sensor + sensor head + FPGA or similar to have the camera.

[Farhad Towfiq]

Position accuracy is critical for the sensor. Without absolute perpendicularity of the sensor to the optical access the camera will not work for short depth of focus as you will have focus on one side and not on the other side. You must have laser autocollimator to adjust position accuracy to several microns for any quality camera. Protecting sensor from noise is even more difficult and only an art. Any noise on the power supply feeding the sensor will show in the images.

[Jose A. Garcia]

Yes, that's another problem...

But again, the main problem is not the case, the main problem is getting the camera to work.

[John Wyatt]

“This is just a case. We'd still need either an APS-C or a full35mm sensor + sensor head + FPGA” -- I agree. It depends how far back along the process you begin: if you start with a board/sensor, putting it inside an existing camera with a known lens system might be a step forward (though as Farhad says, alignment accuracy might exceed what an amateur can achieve). If you start with a camerahead, you already have all that and will be looking at data writing storage solutions, software, codecs, etc. The way forward for DIY cameras (the best start point) still seems unclear.

I once took apart a cheap webcam and put the sensor inside a 35mm SLR, but I couldn't get a good picture!

[Klas Persson]

Originally Posted by Farhad Towfiq

Position accuracy is critical for the sensor. Without absolute perpendicularity of the sensor to the optical access the camera will not work for short depth of focus as you will have focus on one side and not on the other side.

Though by placing it on the pressure plate it would be on the exactly same place as the film frame would. So the accuracy is already done for you in a way.
Though the sensor is a bit thicker than film i guess so you're right, there would probably be some adjustments needed to get the right focal plane. But that could be done by measuring and checking where your focus is if you could do some smart micro collimation, like on the Brevis.

As for still photo adapters, they do exist. Not for your regular nikon but Hasselblad has some digital backs that you can attach to your old film cameras.

[Aapo Tahkola]

I'm considering having a shot at this with MT9P401 sensor and cy7c68013a usb
2.0 chip. From what I've read from cy7c68013a datasheet, it should be possible
to plug it directly to MT9P401(using gpif), so there is really no need for fpga.
This chip runs at 48 MHz so 1440*1080@30fps is max.
Would be nice to know if MT9P401 is pin compatible with MT9P031...

You can't use huffyuv
or similar algorithms (effectively) unless you perform debayering, so I think it's
probably better to work on bayerized data.
So I have been considering
http://www.eurasip.org/Proceedings/E...rs/a5p-j05.pdf /
http://ltswww.epfl.ch/ltsftp/ICIP2007/pdfs/0200353.pdf .
Problem with this is that it's quite computationally heavy algorithm at
these data rates. I'm also little worried that blackfin might not have
enough wide io for sram to run this kind of compression. Rates are 253.68 MB/s
for blackfin with 16 bit bus clocked at 133 MHz and 381.47 MB/s for TMS320C6413
with 32 bit bus clocked at 100 MHz(max allowed). It probably takes something
like 5 times the input data rate to run this kind of compression algorithm.

Also, I do not yet know whether it's computationally possible to do this in real-time. TMS320C6413
at max speed would give 32 MMACS per pixel for 1920*1080@30fps assuming all
operations work on 16 bit data. Double that for 8 bit data.
Compression ratio is about 0.5 - 0.6 on "average" case. Probably
something like 0.3 - 0.4 if compressing delta between predicted and past frames with this
algo. I haven't yet implemented the ffmpeg codec so I don't know what
the exact compression ratios would be for video.
The big problem with lossless compression is that compression cannot be guaranteed
if the scene is just too complex.

Final version would probably have:
-some fast dsp
-sony psp screen
-arm7 cpu
-sata/ide hd
-http://www.dealextreme.com/details.dx/sku.7533

It's very unlikely final version becomes reality though.

[Jose A. Garcia]

Hey, that sounds really good!

I have a few questions:

- Why 30fps? If we're trying to build a cinema camera we'd need 24fps.
- Are other resolutions possible? You talk about 1440x1080. What about 1920x800@24fps?
- Is this a pc-less camera?

- Lossless compression would be good but it's not a must. Low lossy compression is ok too if it stays between 4:1 and 8:1. New versions of Cineform are 4:1 and the RED One compresses even more. What about Motion JPEG2000? And visually lossless RAW?

- Are you considering sharing the cy7c68013a code you write with the community, so everyone can buy all the parts needed and build a working cam?

- What can I do to make your life easier?

Thanks a lot. It really looks like we're finally getting somewhere.

[Aapo Tahkola]

1920*800*24/1000000 would be 36.864 MHz which is fine as well. In theory, any resolution and fps with w*h*fps/1000000 < 48 should be fine. Bits per pixel is limited by transfering speed of usb 2.0. I'm not sure how advanced cy7c68013a is so only 8 bpp and 16 bpp(remaining ignored) could be possible.
First version having cy7c68013a would plug to usb port. Second version would probably have cy7c68013a as well.
cy7c68013a is the same cypress FX2 Alex Stewart was talking about earlier.

According to that paper, JPEG2000 compression ratio isn't as good as with this algo. I'm not familiar with JPEG2000 but I'd think it's also computationally more intensive than this. One candidate for near lossy compression could be
http://ieeexplore.ieee.org/Xplore/lo...418.pdf?temp=x . I do not have access to ieee so I couldn't tell.
It might be possible to come up with a good(fast) lossy algo by taking some of ideas proposed in papers I posted earlier.
It's not the compression ratio thats the big problem here, its finding a algo that can achieve ~0.5(fixed) compression ratio using low cost and easily available components. More simpler compression algo means there will be less compression. TMS320C6413 costs about 25 euro, which is about as high I'm willing to go.

Anyway, building a standalone compressing version is way more difficult and time consuming than just plugging a sensor to pc via usb. I'd hate to be realistic here, but this project will probably die once the usb version is done.
Compressing version would require team work and a plan that everyone interested agrees to. Alex Stewart's version is too complex and expensive for me.
MT9P401 pin layout is my main concern now.

BTW, Alex Stewart got picture.

[Aapo Tahkola]

Had another look at dsps and found TMS320DM6441. It's little faster than TMS320C6413 and has arm9 cpu, usb 2.0, ata, 100 mb emac, and ddr2 interface. Also has all sorts of jingle bells meant for video processing. At 30 euro it seems like a good candidate. This one has 0.8mm ball pitch while TMS320C6413 has 1.0mm. Ball layout looks much worse than TMS320C6413's.