4:4:4 10bit single CMOS HD project

[Wayne Morellini]

<<<-- Originally posted by Rob Lohman :
Wayne: I did my first programming in low-level assembler before
moving to C(++) and then on to the Windows platform. I'm pretty
sure I know exactly how a PC works internally and how Windows
works as well. I've written assembler boot-loaders and some
low-level Windows stuff. The only thing I ain't really good at is
anything with Unix/Linux on the PC. Oh well...
-->>>

Well your two levels up on most other app programmers, most of which wouldn't realise much about these timming issues, let alone how to handle these hardware timming issues, as ussually only highend 3D games need this much performance. But I thought it might help and provide nice reading for everybody else. I'll delete it then, no use cluttering up the space.

Thanks

Wayne.

[Les Dit]

Thanks for the two frames. I might try the variable gradient software on it and post the color result. That seems to be the best demosaiking method right now.

Can you post two raw 10 bit black and whites to allow me to evaluate the noise component? That would be frame N and frame N+1 from a capture.

-Les




<<<-- Originally posted by Obin Olson : 16bit tiff file

http://www.dv3productions.com/test_i...-24fps1835.tif

pre-Bayer filter black and white image:

http://www.dv3productions.com/test_images/pre-bayer.tif -->>>

[Les Dit]

Just for 5hits and giggles, here is a few seconds of media9 compressed output originating from my JVC HD10. Of course it is hampered by the 8 bits of color depth, and it's non-manual modes, etc, but those who are used to looking at DV footage should take a look.
This camera is a single chip 1.1 Mega Pixel sensor. It records to a 19 megabit mpeg2 stream. About $4K for this camera.
The detail is good, look at the thread details in the backpack label,
they seem to have survived the media9 compression!
The file is about 13 meg, it's a 5 megabit data rate.

I shot this yesterday in a park.

http://s95439504.onlinehome.us/park.wmv

-Les

[Rob Lohman]

I must say I'm pretty disappointed with the rolling shutter
effect. It is quite noticable, especially lateron on the movie
when you pan pretty fast along the fence, Obin.

Obin: could you perhaps do some testing with smaller resolution
files?

Capture 640x480 at 8 bit and do this at 24, 48 and 60 fps
(Your computer should be able to handle this in the lower
resolution) and do some similar sweeps along side the fence.

This can show us how much the effect changes at higher
framerates.

Thanks for testing!

[Obin Olson]

Rob it has nothing to do with fps it's all in the mhz of the camera sensor...if you shoot high enough mhz it goes away..well almost goes away

[Rob Scott]

Obin Olson wrote:
it has nothing to do with fps it's all in the mhz of the camera sensor...if you shoot high enough mhz it goes away..well almost goes away

Hmmmm ... I wonder if it would be possible to capture at a higher frame rate and *average* two frames together in real-time? If done properly, it would reduce the rolling shutter effect and -- I would think -- simply produce more motion blur.

(This method would also work on a very-high-frame-rate camera like the other Micron chips we've discussed. Of course, at a 450 fps frame rate we'd have to do a lot of real-time averaging! But it would be an integer operation, assuming you're doing it pixel-by-pixel ...)

[Jason Rodriguez]

Hey Obin,

Just curious, how come there's no motion blur on any of your footage? It looks as though everything is shot like the opening of "Saving Private Ryan". Especially the shots that show the rolling shutter artifact, it seems like the main reason they're even apparent is because there's no motion blur to hide them.

Again, I'm wondering, since SLR's, digital cameras, etc. have electronic shutters that have chips with rolling shutters, how come they don't exhibit these motion artifacts?

[Obin Olson]

because the shutter speed is high, it's outdoor light have to put the shutter speed high to get a nice DOF

[Rob Scott]

Obin Olson wrote:
because the shutter speed is high, it's outdoor light have to put the shutter speed high to get a nice DOF

So, you could reduce the "Private Ryan" effect by using a slower lens or using a lens with an iris?

[Eliot Mack]

I added a 'Platforms' category to the Obscuracam Wiki under Hardware.

I spent some time looking at potential processor solutions. We are processing a very large amount of data, which places us at the top end of the embedded market. I would like to have the system be at least somewhat portable, with an ability to run on battery power even if for a short time.

Opus Solutions is working on an automotive mount chassis for a Mini ITX motherboard that looks like a perfect fit for our needs. It has a built in high-efficiency DC-DC converter so power is not wasted, and is ruggedized to handle shock and vibration found in cars.

http://www.obscuracam.com/wiki/wiki/...Page=Platforms

RAID is really hard to find in a portable system. The motherboard on there has support for 2 channels of IDE, but I don't know if it can be striped to RAID 0--I'm checking on this. Looking at Rob's calculations from a couple messages back, it appears that we can capture a 1280x720 24 fps stream with a single drive, but not 48 fps. There will have to be some experimentation here, but I can see a situation where a laptop/lower power drive is used for portable applications and the big RAID is used for studio applications where high frame rates will be used.

Eliot

[Les Dit]

Camcorder users, as I'm sure Obin knows, use ND filters to keep the Fstop open for less DOF.

Throwing a few stops of ND in front of that lens would slow the shutter way down to avoid the short shutter angle look.
But the rolling shutter would look much much worse.
Until camera firmware is developed that allows longer integration time with a fast readout, the camera will only be good for that 'Saving Pvt. Ryan' shows.


Is that firmware being worked on, Steve?

-Les


<<<-- Originally posted by Rob Scott : So, you could reduce the "Private Ryan" effect by using a slower lens or using a lens with an iris? -->>>

[Obin Olson]

I am not sure your right about that..I think with ND you can still keep your mhz fast..I am only using hugh shutter speed so that the image is not blown out in the highlights

[Steve Nordhauser]

OK, maybe I need some education here, but this is what I think. There are a number of ways to control the light integrated by the sensor. All of these can be independent of the frame readout time which should be the only variable that matters with rolling shutter artifacts.

First, the sensor has an electronic exposure control. The rolling shutter cycle is read, reset and start integrating for each line. What maybe I didn't say was that if you shorten the exposure time, there is a delay before the line is reset. This means that you can reset the line as short as one line prior to readout giving very little integration time. This does *not* change the rolling shutter artifact because it didn't change the readout time - time from the top line to the bottom. I suppose this is the equivalent of shutter speed since less integration will have less blur, but digital camera people call it exposure control.

Next, you get the iris on the lens, certainly influencing the DOF. Then you get NDFs which I don't think influence the DOF but only give the same effect as the electronic exposure control without the change in exposure.

You also have analog gain controls within the camera that let you integrate for long times but set the gain to a minimum. I think that if you start with a NDF and some gain, decreasing gain is like adding a NDF.

Now that I think about it, if you have separate color analog gain controls within the camera like the SI-1300 does, by changing them inidvidually, you are applying different color filters prior to digitizing. This means that you still have the dynamic range in that color as opposed to applying them in a post process step. I don't think you need color optical filters at all.

[Les Dit]

Photons are photons.
If you want to have motion blur, it means keeping the sensor 'on' for a longer period of time. To do this longer exposure, or 'integration' as the electronic cam people call it, you have to limit the number of photons hitting the pixel well to avoid overload ( blowout, clipping,etc )
There are only three ways to do this.
1> reduce the photons with an ND
2> reduce the photons by closing the aperture down
3> Darken the scene

Film people like short DOF. That leaves us with method 1 and 3.
Method 1, the ND filter is the easiest. Unless you can ND the sun!

-Les

[Les Dit]

Steve,
Turning gain down works fine, as long as the photo site ( well) has not been saturated ( blown out , in photo terms ).
If too many photons have hit the sensor, you have lost image information, you can't bring that back with a gain setting.
I just wanted to clear that up.

So where are the two 10 bit consecutive frames?
Is the noise on this cam that bad that it's too embarrassing to post?
Is this really an 8 bit camera, for all practical purposes?

-Les