New DIY HD Cinema Camera Project

[Jose A. Garcia]

Too bad... We need constant frame rate for this. Damn! Almost there! 1920x720 at 30fps!!

[Jose A. Garcia]

Hey! Did you think I was going to let you all without any clip from the micron board?

http://www.cus-cus.net/dvinfo/Video03-1280x533.avi

The three clips are 1280x533 (1:2.40) 24fps. This of course is me. Please don't pay attention to that and focus on the camera. This was recorded at about 2 in the afternoon. No color correction. In fact the camera has a white balance and color correction option but these three clips are just what the camera actually sees. Debayered, Mjpeg encoded and uploaded.

http://www.cus-cus.net/dvinfo/Video04-1280x533.avi

Very slow pan. It's not smooth, I know :). It was also 8 in the evening so it was a bit dark. A bit of noise too.

http://www.cus-cus.net/dvinfo/Video05-1280x533.avi

My hand. More darkness. More noise. I'll record something better tomorrow.

So... As I said, I'm looking for another solution. In fact I'm looking for two different solutions:

- One would be a 1" sensor (or 36x24mm to match exactly a 35mm frame) able to deliver 1280x720p at 24fps. In fact the Pike camera (with a 1" ccd) does it and delivers full HD but it's $5000

- The other one would be a smaller chip (about 1/2.5" or 1/2"... or even 1/3") but able to deliver 2k or full HD at 24fps so I can add an adapter.

If anyone knows about something like that...

[Take Vos]

Hello Jose,

Your clips show good promise, I didn't find it too noisy.

I myself am working with the Pike.

It does not handle full frame HD @ 24fps. The firewire bandwidth is the limitation. Currently It can handle 1800 x 750 @ 24fps, 14 bit (transfered as 16 bits) pixels. It will do full HD at 8 bit per pixel I guess.

I have heard that they will release new firmware this month that does packed 12 bit transfers. So that it will do 2048 * 854 @ 24fps, 12 bit.
There is a Pike that has a 2048 x 2048 1.2" sensor, I myself have the 1920 x 1080 1" sensor.

Currently my Pike has a noise problem that I am not yet investigating. I am confident that I can get rid of it somehow (cooling, battery power, electrical isolation). But I need to get the software of high enough quality first.

[Jose A. Garcia]

I supose the 1/2" version will cost more than $2000 too and it will also have those noise problems you mention. I thought CCDs were better than CMOS handling noise.

[David Delaney]

Wow, I thought the video looked great. Nice job!

[Take Vos]

Hello Jose,

I think CMOS actually handles noise better as it can handle much lower light conditions. It only becomes a problem with very large shutter times (30 minutes) and very low light like in telescopes.

But I am not sure what is going on in my camera. First there is some static noise (non uniformity of the pixels), strange thing is it doesn't seem to go away with a simple flat field calibration.

Also there is some weird non linearity going on, which shows up in the waveform as dark stripes. At first I thought it was an error in my part, or some weird LUT (which are disabled), but they are different from the left and right part (the sensor is divided through the center) which means the non linearity is caused before the AD converter. It seems to only show in the dark part of the image, so it could be the non linearity of the transistors in the amplifier, maybe with the bias I can skip above this non linearity part.

All-in-all, there is a lot to figure out.

Cheers,
Take

[Jose A. Garcia]

I've just been contacted by Omnivision. They'll send me a demo board in a few days.

Take, I know this can be too obvious but, have you thought about the possibility of your camera being damaged?

[Take Vos]

Hello Jose,

It could be a possibility, I have thought of this, but I thought that normal sensor data would look pretty bad without calibration.

I will have to contact the support department with some of the images I have taken.

Cheers,
Take

[Jose A. Garcia]

Well... You can see it's not true. As I said, the clips I posted were completely untouched. I didn't even use the internal white balance included with the camera.

I've also seen images taken with other Alliedvision cameras and they don't look bad at all.

[Jose A. Garcia]

Hey... Alliedvision has the Marlin F-131C with a 2/3" GLOBAL SHUTTER color CMOS sensor delivering 720P at 25fps. That's the sensor size of the SI-2k Mini. It's not 2k, but it's HD, global shutter and it doesn't need an adaptor to get a very nice DOF. You can add a C to F mount converter if you want.

It also has less pixels than the SI-2k with the same size. That means more light, less noise... and for 900euro!

The only bad point I see is that it doesn't deliver full HD.

[Solomon Chase]

Originally Posted by Jose A. Garcia

Hey... Alliedvision has the Marlin F-131C with a 2/3" GLOBAL SHUTTER color CMOS sensor delivering 720P at 25fps. That's the sensor size of the SI-2k Mini. It's not 2k, but it's HD, global shutter and it doesn't need an adaptor to get a very nice DOF. You can add a C to F mount converter if you want.

It also has less pixels than the SI-2k with the same size. That means more light, less noise... and for 900euro!

The only bad point I see is that it doesn't deliver full HD.

The Marlin F-131C uses the ibis-5a sensor. Search the board for info. Ibis-5a is an "industrial type" sensor used by sumix, avt, silicon imaging, pixelink, etc. in some of there cameras.

Not very good color performance, but global shutter and shallow DOF are a plus.

[Jamie Varney]

Originally Posted by Solomon Chase

The Marlin F-131C uses the ibis-5a sensor. Search the board for info. Ibis-5a is an "industrial type" sensor used by sumix, avt, silicon imaging, pixelink, etc. in some of there cameras.

Not very good color performance, but global shutter and shallow DOF are a plus.

Anyone know if the ibis-5a sensor itself can be purchased in single quantities?

Edit: Never mind, I guess digikey sells them.

[Wayne Morellini]

Originally Posted by Take Vos

Hello Jose,

I think CMOS actually handles noise better as it can handle much lower light conditions. It only becomes a problem with very large shutter times (30 minutes) and very low light like in telescopes.

But I am not sure what is going on in my camera. First there is some static noise (non uniformity of the pixels), strange thing is it doesn't seem to go away with a simple flat field calibration.

Also there is some weird non linearity going on, which shows up in the waveform as dark stripes. At first I thought it was an error in my part, or some weird LUT (which are disabled), but they are different from the left and right part (the sensor is divided through the center) which means the non linearity is caused before the AD converter. It seems to only show in the dark part of the image, so it could be the non linearity of the transistors in the amplifier, maybe with the bias I can skip above this non linearity part.

All-in-all, there is a lot to figure out.

Cheers,
Take

You are using CMOS? It partly sounds like the CCD split sensor window CCD scanning problem, where because of technology speed restrictions, they use different AD converters for the two halves of the sensor. To get around that problem they simply calibrate them. The rest sounds partly similar to what we have seen on Sanyo HD1 in dark noise, and fix pattern noise (which is removed with fix pattern noise removal). I suggest contacting the supplier and ask them about the problems.

[Wayne Morellini]

Originally Posted by Jose A. Garcia

Wayne, I know you've been researching on this for quite a long time. I also know you were looking for a complete independent solution but I'm just looking for something to plug to the computer. Is there anything out there with good image quality and global shutter (if possible) that's able to deliver 2k 1:2.40 or 1920x800 at 24fps? I don't care if it's GigE, CameraLink... As long as the whole package's not more than $1,300-$1,500.

Thanks!

I stopped too long ago with the research. Go through my thread technical thread and you will find links to lists of manufacturers and cameras. By sensor, Micron had some sort of global reset mechanism that helps. Cypress Ibis and Luma (was that the name) had global shutter. Some Kodak had global shutter, and I don't remember much apart from that. Altasens has an heap of new sensors. Sumix and some other manufacturers do cheap Ibis cameras, and Elphel and some others Micron.

Now, here is the rub with the Ibis5a cameras, do it wrong and you land up with an worse sensor, do it right and you land up with an better camera, problem is many seem to do it cheaply, without good external Analogue to Digital converter, and other external support circuits they put in the Drake version.

But the problem with the research, was that I was in too much ill health to do the research for everybody, who were more than prepared to be erratic and not do much research. Even getting good camera companies to do the simple software changes to enable video streaming and control would have been good. If we had all gotten together and done the research consistently we would have been TEN times further on, cheaper price/better camera. Sumix was going to make an cheap camera for us, but we let them down and ran after Silicon Imaging, Micron and Altasens, that only delayed us for years and gave us cameras at 4-6 times the price, and Sumix let us down and went away. Now we have Elphel, and I have got to admit, I am less than cinematically appealed to it, it is community only, and everybody keeps disappearing from it and alternative projects keep turning up. We will get what we deserve. We really need to pull together to get an company to support us.

Having said all this, here are the simple options, under $500 webcam+optical adaptor. Under $2K Elphel, or HD camera plus Intensity HDMI/Component. Red is coming out with an cheaper camera. As Sumix is no longer interested, why not ask Silicon Imaging about an cheap Ibis5a/new 1/2inch Altasens/Kodak/hopefully not Micron 720p, camera based on their cineform Digital Cinema POV, for $1000+ (you supply computer/hard drives). Such an camera, based on an improved version of their Ibis5 camera, can't hope to compete with the top end Altasens, so they can afford to cut us an break, and sell it as an industrial camera too. Of course, it could be GigE direct to hard disk. The Ibis and I think, new altasens, offers latitude extension technologies, however, it is preferable to be able to do at least 50fps 100% shutter for transfer between different media formats, which is where the Ibis and many others fall over.

[Wayne Morellini]

Originally Posted by Igor Babic

Wayne, this is what my friend got from sumix support team:

"Re: [sumix.com #8710] Question regarding SMX-M72 camera
Dear
Good news.
We almost finished testing the fast recording mode. According to prelimenary experiments it allows to write app 30-32 frames per sec directly to HDD at resolution 1920x720. The recording is limited only by HDD capacity. The recorded file contains some simple header and sequence of RAW (bayer) frames.
We need to prepare new tools for conversion this format to avi or to a set of separate files (bmp, jpg...)"
..
"Direct recording to HDD can not getting on smoothly. From time to time there will be jerks and stops. The effect is due to Windows and incompatibility of data transfer speed of USB2 and HDD"
I dont know if this applys to your camera...

This USB problem, is what they were supposed to be going to solve, around 3 years ago. Performance depends on main board with good USB interface, windows configured for realtime, alternative USB interface driver, and efficient third party windows realtime core. With some, or all, these factors they should be able to easily get stable performance. However, trying to get it to work on all systems, without any enhancements is probably the problem. This doesn't matter so much, as long as we can get an range of regular hardware to work properly with the software. You could ask them about this.