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

[Wayne Morellini]

Now that we seem to have found most things needed to specifiy a potential camera design, I thought I wouild list them out (haven't been keeping up with the Wiki, but you might like to list this Rob).

Re-edited: cheap battery info added.

Camera 1: Simple Handycam

- camera 720p single chip
(variations allow higher specs)
- Single Hard Disk drive
- small form factor motherboard (nanoITX series, embbeded PC's possbile but expensive).
- PC Interface USB/Ethernet or higher.
- Small form factor or custom case
(At the moemnt Oblin is using an exisitng 16mm Russian Movie camera)

In the future multiple drives and chip versions at higher resolutions would be possible.

My thoughts: if you look at the Hoojum Nanode cases www.hoojum.com/index.php , posted earlier, they look like oblong, tin cans (there is a tin opener that will cleanly cut the can along the rim itself). Maybe we could find non ribbed oblong cans and do the same, or even paint some oblong plastic containers metalic, for ammusement, and also to get our footage straight into the can ;) One thought is to get mouldable mirror perspex and shape one for handheld use (and add some case mod lighting).

Camera 2: Straight shoulder mount

- camera upto SHD multichip (depends on the variations, interface, processing and drives involved)
- Multiple Hard Disk drive.
- Small form factor motherboard (miniITX series, also mini atx, btx, embbeded PC's possbile but expensive).
- PC Interface Cameralink, 10/Giga Ethernet, HDMI (DVI), USB3/Ext PCI express.
- Small form factor or custom case

Each camera also has:

- External LCD (or touch sensitive) or head mounted display.
- Custom control panel or touch plate, and/or, firwire, or standard external controls.
- Standard lense mounts to suit sensor chip, others through adaptors.
- Adpators through image reduction (brighter image) and projection (better DOF).
- Battery, buy your own, but I prefer to get a cheap battery case and stuff it full of cheap nicad cells. Not as good as li-ion but if you know where to get hi-densiity chineese cells cheap off an distibutor, cheap to repack.


Additonals:

Camera: Any supported camera.

This is being achieved by either custom support, plugin or profile files. The sort of support will change during developement.

Software: Custom Capture (the Rob Mobile) and any supported Editor. Or alternative commercial product.

Format: Any supported format

This is being achieved by use of standard (or if needed, custom) codecs and plugins for the capture software and the NLE Editor.

Variations:

Variation 1:

External capture/compression card to take from high speed camera interface to low speed PC interface. Allows lower CPU use, and by passes Cameralink PCI speed restriction.
- Use of Clearspeed like parrallel processor or FPGA.
- But codec support is restricted. So good codecs need to be picked for lossless, visually lossless, and high comrpession are needed, preferably open unlicensed. With Clearspeed this canbe reprogrammed.

Variation 2:

PC variation of variation 1: Using PC with inbuilt 2.3GB/s Cameralink interface (by passes Cameralink PCI speed restriction). If a comrpession engine is added it then reduces cpu load. Main problem is that the PC is tide loacked into a maxioum capability, and gets outdated. But eventually PC speed will outstrip camera SHD/UHD requirements and it will no longer be a problem. Also this puts more of the system in the hands of low volume manufacturers who can charge what they want.
- Extra processing through Clearspeed like parrallel processor or FPGA. With Windows using Clearspeed as a coprocessor resource, a standard codec could benefit (Sometime ago MS put API's/Device drivers in to allow processing requests to be passed to DSP's (simular toi Direct X). I do not know if they are still supported. Advanced stuff.

Variation 3:

Variation 1 pluss ATA drive storage.

Variation 4:

Variation 1 or 3 in the camera head.

HDMI is compatiable with DVI used on AGP graphics cards (but which have HD digital capture is unknown) so the bandwidth restrictions are removed.

So does this look like a good overveiw of the possibilities and technology?

Cameras with 10 Gigabit Erthernet or HDMI are the way to go (hint, hint). If 10 Gigabit Ethernet is going to be commonly available in the next year or two. I think that USB3, PCI Express External, and SATA 300 are good for the next level down (except we need all the SATA channels). We could even buy 8MP today and use it in HD through HDMI till a good enough computer becomes available.




Steve, I just reread your post, and realised that was a BBC wavelet codec, we should defintely watch that one, if it turns out to be near the best it might be good.

We have a number of open codecs being mentioned here, could I suggest we make a list of the industry leaders for Lossless, Near Lossless, Visually Lossless, medium and high compression, and bayer, and see if any of these free codecs come close in quality (compression for lossless) and computation speed.

About the firehose, it is like sucking on a firtehose for hours doing the resarch too.

[Rob LaPoint]

Hi everybody, I am obviously new to the forum, although I have been reading over everything for the past couple of days. I am happy to see some people finally taking the bull by the horns and trying to make their own solutions. This is the way of thinking that I have always lived my life by. I have been toying with the idea of building my own camera, but after reading everything on this forum I have decided that it would be best to just join the ranks of you guys. Unfortunatly I do not have the electrical expertise to really be much help at the stage you guys are at right now. I wanted to re-iterate everything that has happened on this forum just to see if I am understanding everything correctly.

- Various companies like Sumix and Silicon Imaging are working a few camera designs, the best of which is a 3 sensor camera using the Altasens cmos chip that is being used in the Kinetta (although he is only using one chip)

- These cameras are capable of shooting at a variety of frame-rates and they output RAW 12 bit images via some sort of data cable ie: USB2, eithernet, ect...

- A frame grabber of some sort must be used to actually record the data unto disk

- The companies creating the cameras are not building housings to go with them, so that is something that must be built.

- Rob is working on capture software to record and work with the RAW data so that it may be brought into an editor.

- We should start seeing cameras in the next couple months

If any of that is wrong please let me know!

I do have a couple of questions though after reading through this forum and Obins forum.

- Will these cameras have the rolling shutter problems that Obin is running into? Will there be a variable shutter speed?

- Will it be possible to just run a cable from the camera to a suitible computer system for recording rather than trying to have some onboard computer to do the recording?

- What sort of lense options will we have with these cameras?

All in all I am really excited about everything thats happening here. I said before that I am probably not going to be too helpful in the development of this stuff, but I can hopefully contribute to this effort. I am working on a series of computer controlled motion capture systems from a simple pan/tilt head up to a full crane and dolly system. I am hoping to produce these things for far less money than they are currently going for. My thinking is probably the same as yours, I am sick of companies charging film/video producers through the roof just because the big budget productions can afford it.

[David Newman]

<<<-- Originally posted by Wayne Morellini : <<<--
One thing that I would do to get high lossless, would be to eliminate niose/grain becuase it doesn't add anything, detracts from the real image, and is hard to compress. I sit on an awfull lot of patentable ideas around here, thgat I have to just ignore.
-->>>

This is why what you describe isn't lossless. This is exactly what many extreme compressors do to get content over the web. However, the "grain" is part of the image, particularly from film source. This should be preserved, not discarded. Also noise can be your friend (true, not for compression); a small amount of noise helps reduce banding due to the limits of 8bit and even 10. Noise is the natural dithering algorithm. Basically true lossless compression of naturally gathered images is between 2 and 4:1 (the higher the resolution the higher the compression that can be achieved.) Yes, there are special cases like digital black, yet we don't really want to see feature length black, so there is little value in exploiting these cases. Let's stick to technology that exists.

[Wayne Morellini]

This is one of the only compromises I'm prepared to take to get higher lossless compression. The niose, and the grain are not truelly part of the image, but a defect in the sensor's performance (realising the senor is not perfect anyway). I'm not talking about eliminating texture here, or real detail, only the stuff that the sensor "imagines" it sees, and in such a way to at least partly restore the real image (wether by calculated averaging or interfame comparison).

On the black compression, I was just illustrating what rediculouse compression ratios we can get by using certain images. I'm sorry if you thought I was saying we should depend on that.

[Wayne Morellini]

Rob Lapiont, yes to most of your questions.

Various companies like Sumix and Silicon Imaging are working a few camera designs, the best of which is a 3 sensor camera using the Altasens cmos chip that is being used in the Kinetta (although he is only using one chip)

Maybe, but the 3chip will be after the success of the single chip versions.

- These cameras are capable of shooting at a variety of frame-rates and they output RAW 12 bit images via some sort of data cable ie: USB2, eithernet, ect...

Mainly RAW 10 for the moment (I think the Aktsens will do 12bit). Most 12 bit requires a special double sample according to Steve (I don't know about the Altsens though).

- A frame grabber of some sort must be used to actually record the data unto disk

Not if your using USB cmaera or GB ethernet adaptor, but USb unrealiable.

- The companies creating the cameras are not building housings to go with them, so that is something that must be built.

Yes.

- Rob is working on capture software to record and work with the RAW data so that it may be brought into an editor.

Yes.

- We should start seeing cameras in the next couple months

At the earliest.

I do have a couple of questions though after reading through this forum and Obins forum.

- Will these cameras have the rolling shutter problems that Obin is running into? Will there be a variable shutter speed?

Some don't I think that Altsens might capture andf output at the same time so they will be like global shutters if that is the case.

- Will it be possible to just run a cable from the camera to a suitible computer system for recording rather than trying to have some onboard computer to do the recording?

That is the only way for the moment.

- What sort of lense options will we have with these cameras?

Whatever the manufacturer ships with it, different sized chips have corresponding lense systems, otherwise you will have to use a lense adaptor using either projected image (keeps DOF but sacrifices some light) or condensed image to the new chip (brighter image but same DOF as the chips normal lense). Mounting a lense straight without sizing down the image will result in loose of angle of vision.


Wish you luck and thank yoiu for coming Rob.

Everybody else see you in a couple of days maybe.

Thanks

Wayne.

[Steve Nordhauser]

Rob, Rob, Rob, Wayne and Obin (mind if we call you all 'Rob' to avoid any confusion?).

To elaborate on some of Wayne's answers....The Micron and Altasens chips have integrated A/D converters so at the max bit depth you are stuck with what they provide. Micron is 10 bit, Altasens is 12 bit.

The cameras come in small industrial enclosures and could be used that way on a tripod. They can be (awkwardly) used with existing software (see Obin's workflow) right now.

Altasens is rolling shutter but at up to 60fps. By the way, there are several Altasens chips. There is the ProCam 2560 1.3Mpix, 3520 1920x1080@30fps and the 3560 1920x1080@60fps - those being max rates if the clock is programmable. Our new camera is using the 3560, but I don't think they are commercially available yet.

The cameralink standard allows up to 10m cables. gigabit is 100m. That should be enough to remote the computer.

[Rob LaPoint]

Thanks for the info! I have a couple of questions regarding the rolling shutters because I am just learning about them. The problem with them is that because they 'roll' the exposure over the sensor actually happens at slightly different times and this can cause problems especially when the camera is moving over something like a picket fence because there are many vertical lines. It is my understanding that if the shutter is moving fast enough say at (48 frames a sec with only 24 of those frames being captured) that the problem can be avoided? If that is the case would it be feasible to actually run the altasens 3560 chip at 60 FPS or would you be running into alot of rolling shutter artifacts? Again my understanding of this is very new so If I am off base please forgive me.

Rob L

[Obin Olson]

Thank goodness your using the 3560 Steve, I was going to ask you that question..it's the only chip worth using IMHO

[Obin Olson]

Just to be open and honest this Micron camera does have a CMOS issue..when you shoot a subject that is VERY hot in a small spot like a reflection of a light on glass etc you get some sort of ghosting..it's alot like a vertical smear on a ccd but it's horizontal...when you push the footage in post really far you can see it and it's not good..not sure if I have a bad chip or this is just how CMOS chips are..I know that I have seen really cheap spycamera cmos tv cameras and they have this effect and it's MUCH stronger then the 1300..dunno just a FYI....I am sure the Rockwell chip has fixed this or they would not sell it for HD camera production

[Laurence Maher]

This will seem laymen-ish, I'm sure, but I thought there was a methoed of using a particular shutter degree in order to compensate for the "strobe" effect from a rolling shutter. I guess I'm wrong ont this? (Please reply and describe why/why not and the difference between rolling/global shutter).

Thanks!

[Rob Scott]

Eliot wrote:
I posted some potential Mini ITX power supply and motherboard designs on the Obscuracam Wiki

Thanks for doing that! I think I posted some -- a few of the same ones in fact. When I have a few extra minutes I will try to reorganize the wiki to make it a bit more coherent.

I think the Kinetta design of making the recording module detachable from the camera head is very smart

I like the idea too, but I'm concerned that we'd just end up at the same price point as the Kinetta. I guess my feeling is, if you can put enough capacity in the camera -- 3-5 hours? -- then it should handle a typical shooting day with room to spare. Transfer the data off overnight -- perhaps a customized FTP program to double-check that everything was downloaded properly -- and erase & go in the morning.

OTOH, I would love to see a professional case for this project, dockable or not. Keep me posted on your thoughts!

Steve wrote: mind if we call you all 'Rob' to avoid any confusion?

Ahhh ... G'day Bruce! ... another Python fan in the crowd! :-) At my old job, whenever we did job interviews, I would toss out a few Python quotes during the group sessions just to see who would respond to them. It suprised me how many shot something right back (but I also did get a few funny looks).

[Juan M. M. Fiebelkorn]

I think we need to find a way to use a clasical rotating mechanical shutter, as 35mm does.Is this possible, Nordhauser?
I know of a couple of industrial cameras that have this option.

[Rob LaPoint]

I agree, I have to say that I am so impressed with what is happening on this forum but out of all of the roadblocks the rolling shutter is the only one that worries me. I know that the accepted fix for the 'rolling' artifact is to up the mhz but keep the frame rate low (shoot 48 frames a sec but capture 24) The problem that I have with this is I am worried that it is going to destroy motion blur which is a hallmark of the 'film-look' Even with the highest resolution possible if there is no Mblur its going to have a video look. I know that there are more expensive chips that have global shutters but there has to be another way. Obviously Steve Nordhauser knows alot more about this than I do (since I couldnt have told you what a rolling shutter was a week ago) Obin has also seen the effect of this first hand so if one of you guys has an opinion on how much motion blur you can get with the rolling shutter that would be appreciated.

Also I just noticed but Steve Nordhauser and I only live about an hour from each other, maybe I can drive down when I buy a camera and save on the shipping ;)

[Steve Nordhauser]

I think one of the problems we, as camera manufacturers are facing in bringing a global shutter camera to the DV market is that the sensor companies don't perceive the need. Here are some sensors with global shutter and their specs as they pertain to this group:

Fill Factory - The much maligned IBIS-5 - Can do 1280x720x30fps, 12 bit with an external A/D. 6.7 micron pixels. Color and mono - noisy and the color smearing is pretty bad. Why? Made for machine vision.

Micron MT9V403 - VGA color/mono at 250fps - sports analysis and machine vision. 9.9 micron pixels

Micron MT9M413 - 1280 x 720 @ 500+fps, 12 micron pixels, color and mono. A contender for sure but 10 bit only, 10 tap - 100 I/O lines and expensive for the sensor and interfacing. Made for speed.

Fill Factory LUPA1300 - Very similar to the Micron MT9M413.

Fill Factory LUPA4000 - can do 2Kx1K 30fps, 12 micron pixels, dual slope, 4 tap (I think) sounds exciting????!!!! Are your hopes up yet?????? Mono only. Military and machine vision markets. Might be nice in a 3 chip camera but expensive.

Since I'm having fun discouraging people (OK, I got up too early) we have a prototype that can do 1Kx1K, 12 bit, color/mono, 30fps, 12 micron pixels. Built two cameras and the sensor is no longer available.

There are certainly some CCD solutions out there using interline and frame transfer. People should look at these but they tend to be expensive and slow. These are the areas that CMOS is winning.

[Valeriu Campan]

It seems that the rolling shutter is not a good idea for this project. I am still intrigued by the Lupa 1300 or 4000 cameras. They have a largger sensor (18mm wide), much closer to 35mm academy. I don't see a rosy future for a 2/3 sensor, that is smaller than the 16mm frame and delivers a similar FOV and DOF and very "video". Most of us are interested in a 35mm motion picture feel. The mini35 adapter at best will deliver about 2 stops loss. Probably the home made adaptors using still camera lenses employ a higher loss of light and more production difficulties with focusing and lens consistency. I think we should concentrate on larger sensors with a global shutter that gives proper motion blur. The cost difference will be compensated by savings made during production using less lights, less equipment or time lost with set ups.