Wayne:
Dont you think that a 1/3 sensor having a 5ym pixel size isn't to small for good low light performance?

Right now I am testing some AVT cameras.
The Marlin F-033c has a 1/2 Sony CCD and the cuality is decent for being a SD camera.

The Pike F-210 is an excellent HD camera but the price is over $5000.

http://www.goavt.com/product_pike.asp

Cesar Rubio.

That is the problem, the price is so high, before you even put the recording solution in.

I don't think that 1/3rd inch 5 micron sensors are necessarily too small with technology and performance like this, but the old saying applies, light, light, light. You would not expect much more low light performance than current low end cameras. In more expensive camera with faster lens, 1 Lux would be possible.

Cesar - Being very interested in the AVT cameras as well, I'm dying to see any images they can produce. I've been in touch with their sales dept., but my requests for clips or frames have yet failed to get a response. In your testing, have you produced anything at all that could be posted? Even individual jpegs of frames would be very appreciated.

Good luck with your project!

Better than h264 codec, even lossless
 

I have started a news thread on this new product, called PSI from Even Technologies. Should be of interest to those exploring codecs for their Digital Cinema Camera projects. If you look at the web link, it shows Windows Media 10, some version of Divx, and VP7 (from people that developed vp3 used in Ogg video codec, but not version 8) all outdoing h264 for web streaming. As usual, all these sorts of tests depend on sample and setup used. They mention both software and hardware codecs becoming avail;able. So, unless they have some deal for custom silicon, it might even be an FPGA, which is of interest to some of our members.

http://www.dvinfo.net/conf/showthread.php?t=80652

It is a big fuss, the area of alternative video codecs is underdeveloped in performance, and few are standouts. Some alternative codecs offer similar performance at much lower processing cost.

I've taken a few steps toward a machine-vision solution, but my knowledge is so rudimentary, I don't feel I'm ready to enter the discussion, even with questions. My little firewire Point Grey Flea was so plug-and-play easy, I think it got me a bit initially overconfident. Recently I acquired a bayer 2/3" camera, but it's LVDS, and suddenly me and my Mac are a ball in tall weeds.

here is some interesting posts!
 

http://www.cinematography.com/forum2...howtopic=19400


J.G. Beckman

Cesar has some AVT pictures on his site...

Wayne, in your vast research on CMOS sensors from a variety of manufacturers - have you come across any offering 1" inch sensors at all?

Back from break.

Has this thread been clipped? I would not mind knowing what that forum was, a good reference. I think an custom made/Digital cinema camera sub forum would be good here, there certainly have been enough threads to justify it.

Dennis,

Yes there are a number of 1 inch sensors. I forget where, apart from Kodak (CCD). It would be worth getting onto the sites of good sensor manufacturers and check, try Cirrus. I can't remember, but try Foveon as well, their sensors are starting to come out in box sensors. I mention this because X3 sensors are starting to hit the industrial sensor market (as posted previously).

Complete small form computer with HDD interface and camera
 

Its not much, but its like elphel 353.

http://www.matrix-vision.com/product...-m.php?lang=en

Wayne - to your knowledge has anyone here built their own HD camera using a 1" ccd? This would be a very interesting prospect.

Igor, nice, what price?

Dennis, around here I can't remember, but I think there are a number of commercial machine vision cameras out there, it is a matter or searching. One place is the official website for the camera-link interface, that has an product search engine. Once you find some you can see if they have them for any other interface. I'm pretty sure I have listed it in this thread, with other lists of products. Come to think of it, have a look at the camera frame grabber packages from Epix, I think there might be one there, or at least an Kodak sensor.

Wayne, I dont know, will ask.

Here is some good links for MV hardware & cameras
http://vsd.pennnet.com/home.cfm

and here for prices
http://www.i-cubeinc.com/cameras-chart.htm

all Kodak 1" have price over 5000$

Epix use Micron and thats why they are about 1000$ for camera&grabber

Imperx & Pulnix use Kodak 1" and Imperx have 1920x1080 GigE version for 5245$ (my favorite, but too pricey). Ilunis looks great also.

I had a look.

There is an Kodak, but frame rate is low, and there is an Zoran.

http://www.epixinc.com/products/sv1310.htm

This is the only one with non micron sensor that has an appreciable frame rate:
http://www.epixinc.com/products/sv1281.htm

Sounds suspiciously familiar.

In the past, one of the problems has been that only an few 24fps capable cmos sensors were being sold in low cost cameras. But time has moved on, and there is presumably more sensors from different manufacturers that will do this, and it would be good if somebody could review whats available and post back?

Now, I would still say it is probably better for people to look at programming HD webcams, Mobiles, and Ultra-mobile PC's for HD video for an cheaper solution. An number of these might contain Micron quality sensors. The other problem is that some mainstream cameras now contain very nice sensors, and you can capture from them. Although these cameras can never match the bayer quality that we can capture, it is close enough to ask about the price of current machine vision solutions.

Forgot to mention:
I think the thing with the Epix cards was, that the free software did not record continuous live motion capture directly and continuously to disk. Even though the package is reasonably cheap, you then have to pay for the more expensive software version to get this feature. If people were interested it might be worth asking them about an special edition for Indies with motion capture.

Conservatively, look over in the main Elphel 333 thread for the projects based on the 353 camera, this is probably the cheapest boxed camera solution.

Those german guys are on to something...

http://www.kamera.abs-jena.de/produk...erentw_de.html

For those that dont undrestand german it sayz that thay have under development 5mpix camera with DVI and HDSDI out at HD video standard resolutions. It will have onboard processing and OSD info. Sound like something red....

BTW.
Has anybody seen this
http://www.arecontvision.com/products_3100.html
http://www.arecontvision.com/products_2100.html
http://www.arecontvision.com/products_100.html
They are like elphel also and are under 700$. Its surveillance stuff that looks like DV on steroids but worth looking...
There is lots of pictures and footage examples. I wish that every MV camera manufacturer have site like that.

Hmm, I need to look at these some over time, but looks like an interesting find.

But this is what I mean, if somebody had the time to look through all the sensors, all the security cameras, all the web cams, you might find more like this.

Thanks

Wayne.

What we need is a 1" sensor so we can get close to 24x18mm and no need for an imaging screen!

New, 4*4*8cm 1080p60 camera, and ultra thin reflected lens system
 

http://www.ue.dk/nyhedsarkiv/10182.aspx
http://www.physorg.com/news89398752.html

I have posted these, because they appear significant, for what can be done in the future.

I am curious what price this camera will be, and compression/data rate, as it talks of network.

The lens is very thin, it works by bouncing light around inside the lens, off of reflective surfaces, until it finds it's way to the sensor chip.

here is direct link
http://www.iis.fraunhofer.de/fhg/iis...bs/microHD.jsp
I have send them email to give us some more info about senzor they using, and about the price.
I think that lan is only for configuration, and preview, HDSDI is their main output

Have anyone seen this:
http://www.ue.dk/nyhedsarkiv/10182.aspx

Would this work for a HD homemade camera?

I hope so.

here is direct link
http://www.iis.fraunhofer.de/fhg/iis...bs/microHD.jsp
I have send them email to give us some more info about senzor they using, and about the price.
I think that lan is only for configuration, and preview, HDSDI is their main output

That link didn't work for me. I did a search on their site and found this:

http://www.iis.fraunhofer.de/fhg/iis...bs/microHD.jsp

Today I have receive answer to my email.
They redirected me to this site:

Please contact.... ( I have edited this part because of personal email adress)
www.easylooksystem.com

I will ask this other person for details about this camera, and to join us on this forum.

Today I have receive email from easylooksystem. They are very interested to yoin us here, and i hope very soon we will find out more about their camera system.

Here it is, prices and basic info. Definetly not for home made guys. Broadcasters are probably they target market. Real live HDSDI video is what this camera gives out. No RAW. Its 18,500€.

Approx $50,000 USD? Is there something wrong with them? You can buy 3 red cameras for that price.

Modula HD CAM latest info
 

Last couple of days I have receive some more info about this camera:

"The MODULA HD CAM is capable of providing RAW data. The RAW data output is not implemented yet, but we are planing to provide 12 (!) bit raw data over HDSDI link and to store it in open exr (http://www.openexr.com/) file format
We are shipping the MODULA HD CAM already. Our customers right now are "early adopters" with special applications. We also got some interest from indie film makers, who are considering to substitute their S 16mm camera with the MODULA. We are trying to get all the info (details, sample clips and stills) you and and other interested parties are asking for and to put it on the new web page. The web page relaunch is scheduled to be ready for the NAB in Las Vegas."

Denis:
18.000€ is about 23.600$
http://finance.yahoo.com/currency/co...submit=Convert

Igor,

I have to ask, the www.arecontvision.com products, they sound very similar to Elphel product, do you know if they are they using Elphel open-source designs?


Thanks

Wayne.

Here is an 2000*850 frame at 85% Jpeg quality, from over at the Elphel thread:

http://www.dvinfo.net/conf/showpost....&postcount=544
http://www.buysmartpc.com/333/333framecc.jpg

Amazing what can be achieved with even an cheap sensor.

Wayne, I will ask them.

Creating new capture software
 

Hello everyone,

A couple of weeks ago I found this forum and decided to take a shot into creating a bayer field recorder software application for Mac OS X. I have already some experience in writing field recorders for audio, see:
http://www.vosgames.nl/

In any case I've looked around for cameras for around a week and it seems like the Pike 210 from Allied Vision Tech is a pretty good contender. As this is a Firewire 800 IIDC camera the maximum resolution at 24 fps and 16 bits (14 bits + 2 bit padding) is around: 1786 x 760 (2.35:1) or 1920 x 654 (2.75:1).

I have also looked at the GigE cameras, but I can not find the specification of the protocol, so I've decided to go for IIDC for now. I've bought a cheap IIDC camera, called fire-i, to start experimenting.

The plan is to let OpenGL do the heavy lifting: dead pixel repair, black image subtraction, pixel sensitivity compensation, and reordering the pixels for better compression. The plan is to lossless compress the bayer image so that it will be possible to record on a single external SATA disk.

The OpenGL pipeline will also be used for previewing: demosaic, color conversion, false color (zebra patterns on steroids), HDR rendering, zooming for focus, more stuff I can think of.

I am not there yet of course, I currently can view (using OpenGL) a grey scale live-image from my fire-i camera.

I've been going round in circles for the last week; should I pre-process the raw data or not. pre-processing would yield better compression, doing these later could potentially give more freedom to choose a different algorithm.

As I already need to process the images in real time to show the camera operator what he is looking at... You can still record unprocessed images by using the NULL calibration settings, which is also the way to capture calibration data.

Cheers,
Take

At last an professional. Take, thanks for turning up and doing this, if you know of other people that might like to do an open camera capture format feel free top invite them. The only professionals we have had here have done commercial products we can't afford (SI and Cineform).

To answer your question on preprocessing, what about both. Some people will want to leave it to last, some will want noise processed out for optimal compression first, others will want it all left to last. I think that, for optimal quality at optimal compression, that pre-processing out the noise is best. For those that want to do it cheaper, pre-processing the colors on the shoot might be desirable (documentary). Cineform allows the colors to be nominated separately from the data (they have lookup table I think) which is an strategy that could also work rather than pre-processing them. AS you are, assumably, using the latest OpenGL GPU acceleration, you should have significant reserves of power to do this, particularly if Apple supports next generation chips in up and coming products.

I would like to point you to an few people that you might like to communicate with: Rob Scott, who is back here developing his own capture program, Keith Wakeham, that has been developing an HDSDI film recorder (with an Bayer mode in mind) and the Elphel 333 camera thread, where we have been discussing a number of bayer compression techniques, including FPGA compression.

http://www.dvinfo.net/conf/showthread.php?t=63677

An interesting side note, Intel has an Direct X 10 part for Ultra Mobile PC's, and some rumor that Apple has an ultra-mobile PC concept. The Intel stuff is coming from Power VR technology.

Wayne.

If you look at machinevision.org (I think) they would probably be able to refer you to GIG-E protocol, or people like Silicon Imaging.

Hey, Steve from Silicon Imaging here. Still keeping an eye on this forum since there is some guts and gore development being talked about.

Take, some thoughts for you. First, watch the rolling shutter artifacts if you run the camera at 24fps. Second, consider your edit workflow. A quick look at the Cineform RAW workflow will show you what I mean.

And lastly, your camera will only be as good as the sensor. On this forum, SI started with Micron sensors which at the time were better than the Cypress/FF IBIS parts. We have moved to Altasens now and there is no going back. The latitude goes from 7-8 stops on the Micron to 11+ on the latest Altasens. True 12 bit capture, low noise. If your goal is a basic camera at a basic cost, a cheap firewire camera can get you going for an amazing price. If you are going to shoot your first feature, do you want to be limited in quality by the capture?

Right now we have set the entry level at $12500 for a Mini with recording software. The edit suite is $2400 with Prospect 2K, Premiere and OnSet. You can use low cost c mount lenses. We are discussing the cash poor indie market - maybe we can announce something at NAB. We will see.....

Sorry, that got sales pitchy - I'm just excited about what we are doing. Last suggestion - don't try to make your software be everything to everyone at first. Pick a resolution, workflow and interface that is doable and focus on completing that.

Steve

Hello everyone,

Thank you all for your comments.

Steve, I can certainly understand your excitement. I seem to want to get in a little bit lower price market than you.

From what I understand the Pike 210 has a progressive CCD chip, it doesn't say anything about a rolling shutter (I thought rolling shutters was only on CMOS).

As for the workflow I think: capture -> openexs(bayer) -> proxy -> editing -> EDL -> openexs(RGB) -> color grading. But I would love to take away the proxy and edit directly in a RGB format, I have to find out how, I have a little bit of experience in writing QuickTime components.

As for CPU time, right now I am reading an 8bit 640 x 480 @ 30 fps greyscale image and displaying it, it takes 1.8 % CPU time.

I am using libdc1394 to capture the image and call glTexSubImage() to update the illumination texturemap. A texturemap in OpenGL can now have any size (not just power of two) and I am using that. I am hoping I can simply send a raw16 bit image to an opengl texturemap without doing any conversion or copying on the CPU.

Next step will be creating a small fragment program to change the picture.

Cheers,
Take

Take,
I checked the Pike data sheet - they don't say what CCD sensor technology so I don't know if you will have issues with shuttering or not - ask them.
I'm glad to see you understand the workflow issues.
If you go to the full sensor size it can be hard to find 1" lenses that don't vignette. Try either a 35mm adapter and SLR lenses (you will get a magnification factor) or these Fujinons:
http://www.fujinoncctv.com/lenses/index5.shtml
Another thing to check is whether, when you go from 8 bit to 12 bit on the data path whether the 12 bit data is packed or sent as 16 bit. Even Firewire 800 will be tight on bandwidth above 8 bit.
Sounds like a great project.
Steve

Hello Steve,

They indeed talk about having slight vignetting, my plan is to capture a 2:35.1 image which will me slightly smaller than the full 1920 width. In the same paragraph they say that you can ask for something other than a C-mount (maybe a direct F-mount?).

Firewire can only send padded 16 bit data, so it is not packed 14 bit data. The Pike has a 14 bit AD converter.

They also allow adding multiple images together to lower the noise, but I have not calculated the internal bandwidth if it would allow such a thing at 24 fps. Of course I have no idea if an image that is build up like this is acceptable for a moving image.

Firewire 800 is indeed pretty tight, in isochronus mode you do not even get the full 800 Mbit/sec, more something like 700 Mbit/sec.

The technical manual you can download from their site shows how to do all the calculations, the FAQ of libdc1394 shows the same calculations. I am not sure how fast the camera reacts on mode changes, but it may allow smooth transitions from over to under cranking.

As for the advice to start simple, that is the same thing I did with Boom Recorder, start with the minimum feature set to record audio. The video recorder may be even easier.

Well, I am off to read the OpenGL orange book (GLSL) and implement a simple bayer decoder without interpolation (grey scale -> red only, green only and blue only pixels)

Cheers,
Take

Hey Take,
Go with c-mount, especially for testing. You can use Nikon F, PL mount, Canon FD lenses and many more with mount adapters. If you special order Pike 210 with f-mount you are stuck with that. You'd also miss out on some cool lenses like the 1-inch F0.85 Fujinon c-mount I have (blazing fast!)

Steve, the AVT Pike 210 uses a kodak KAI-2093 CCD sensor. It has global shutter. It's actually recomended for use in "Video Production".

Nice to see someone new involved here. Looks like you are fairly knowledgable in programming, best of luck in your project.

- Solomon

Also, I would prefer capturing 8-bit with LUT instead of 14-bit. Mainly because of the sensors usable dynamic range, increased framerates, easier post production and some other considerations. 2.35:1 aspect at 14-bit sounds nice though if it is possible.

- Solomon

I think packing would be an special arrangement outside of the standard. Cameras with memory buffering is another way to bring the interface bandwidth requirements down, in this way you can adjust the exposure, readout time without it affecting the data rate.

IBIS will outperform Micron, and also offer many more stops, but only if implemented properly, and not using the internal DAC. But, there are too many Ibis cameras poorly designed or with only some of the refinements.

The two sensor companies I recently mentioned (27 stop latitude, and that high sensitivity through quantum effect) are probably worth looking at (as well as that firewire Foveon X3 one).


Re-edit:
Take, Kodak is an nice sensor company.

I've read the documentation, stacking of two images @ 48 Hz to reduce noise is only possible with this sensor with a maximum of 600 lines. A 1920 x 600 image is 1:3.2 aspect ratio, I think that is a little bit too wide.