To Steve
 

Hi, Steve, thank you for participating in this;

As for outputting R, G, B from a bayer sensor, I think it's much more complicated (I could be wrong, though) than using a single monochrome CMOS chip for the same purposes. Again, I might be proposing something unachievable with this type of technology, but totally utilizing the power of every pixel in an imager would be a huge advantage over 50 + K digital cinema cameras currently on the market. There was a good article in one of the scientific magazines (sorry, can't remember, where I read it) about Martian rover imaging system. They used a similar principle.
The way you could capture R, G, B with only one monochrome chip would be using a digital RGB filter:
http://www.qimaging.com/ccdcameras/sciind/#
http://www.qimaging.com/datasheets/RGBFilter.pdf
The resulting frames could be premultiplied by the corresponding color (R,G,B) on a target capturing system. The same soft responsible for premultiplying would then merge the frames into final stream.
In this scenario, you'd need to triple the number of frames captured on the camera head though - something like 75 fps for a true color representation.
Steve, all this theory is coming straight from pure imagination, - I never experimented with such a configuration; just hungry for pixels :))

Personally, I think 1080p would be the better option if possible. 720 looks good but it's nice to have all the resolution for the big screen.

Hi Steve

Thanks for coming, but it does appear to be a problem as the viper thread is sort of a Sumix project (but they seem to be open as to where they get the sensor from) and this thread was to be a concise list of cameras and improvements rather that technical discussions. But I'll throw it open to this discussion even if I need to open another thread with the camera list.

Most of us are technical novices, so most of the assumptions and technical questions listed in the other two threads would be a good place to start.

About what features, I am interested in inde, documentry, news gathering, and commercals (variety of work). I think one camera can easily do all at a cheap price (compared to a professional HD camera) in higher volumes. I offer the suggestions from the Viper thread, and am interested in your SPIHT adaptive compression routine, as it appears to offer everything from true and near lossless to high compression ratios, which means people could pick what they wanted. If this was included in a independent capture to disk unit (the size of a camcorder body) connected by cameralink, or just the compression routine in the camera head through USN2.0/3.0, then the whole design, and cost, becomes simpler and cheaper. If the capture controller was also a modified Via nano-itx PC reference platform (running at 2Ghz), it could also act as edit unit further reducing cost. A seperate unit could also be used for any camera head.

The number one priority here is quality performance. So for me that includes raw footage (4:4:4 for 3 chip) preferably in true lossless compression form, but for most of us it is the performace of the of the sensor and the optics. We would all love to come close in quality performance to the big boys (your still camera market would feel the same). $5K instead of $50K (though some of us would go for $15-20K). The sensitivity, latitude etc of the camera, and trueness of colours and luminance (apart from all the other image problems), the response curves for the filtered colours, and vlow signal to niose. Maybe you could explain how to read a sensor/cameras data sheet, there are a few unfamiliar measuments in there? With low light performance, it is probably more important for live productions than for inde. I am also worried that using a Bayer filtered single chip will lower this perforamance considerably, and also sensitivity, compared to a 3 chip design, because the filter only transmits some of the light to make the pattern work, is this true? Having said that, a sensor with a feature simular to the latitude of Smal sensor's autobrite technology would be most advantageouse and go a long way to improve single chip performance (not to mention stuff like Foveon's X3 tchnology).

http://www.smalcamera.com/technology.html

For chip size people would like 35mm (though a 35mm lense adaptor is also good), I would even prefer medium format myself, but that is very unlikely. So I would say people would prefer 2/3inch and above (nothing below 1/2 inch as it might be considered a compromise, even with a good 35mm lesne adaptor design).

About super sampling single chip pixels. Is the best way to do it, to sample at each corner of the pixel and in the middle ( to minimise the variations caused by fine features moving withing a pixels zone)?

The resolutions I think people will be most interested in are 720, 1080, and SHD 8mp (pratically in that order). If a single chip could be made good enough, SHD could derive very accurate 720/1080 pixels (because of super sampling etc), and allow the camera to act as a high quality still camera. The higher resoluions are only really practical if compression is available (for true lossless of 5:1 (+), if possible.

For lenses, included adaptors to use any make of cheap secondhand slr manual/auto film lense (for 35mm lense adaptor), and dersired cinematic lenses, would be good.

A top notch 35mm lesne adaptor (if chip is smaller than 35mm), would be good (I think a simple static design could be made good enough). A image flipping mode to correct the orientation of the captured 35mm image would be good.

For the system, most people will prefer to work with nothing but a completely simple to setup and use integrated system. Unfortunately many pople would also like to use their own editors, OS/s, computers and the like. Fortunately most of this could be addressed by using a modular system that captures (if no hardware controller is used, plugin codecs can determine format), and plugins for popular editors and playback software.

Nothing else I can think of for the moment, apart from best wishes and good luck Steve.

Well guys what do you think of that list, anything more?

Thanks

Wayne.

Steve,
The filter idea all works OK as long is it's a still world ( mars , as far as we know ).
Otherwise, you get color fringing when objects move.
-Les

An adapter for still 35mm slr lenses, yes.

OK,
I asked for information. The problem is that not much comes free. There is a threshold where the sensor costs are not the most significant system cost. PCs are getting cheaper but when you start pushing a bus and mass storage above about 60-80MB/sec, there is a cost step. It happens again at around 150MB/sec. Many of the things you want to leave behind - Bayer for 3 sensor, 30fps for 60+fps, 8 bit for 12 bit, large sensor area all increase cost. A 3 chip camera is really 3 semi-independent cameras with a $1-3K prism. 3x the data to store. That is why Bayer is so popular. If you use a good algorithm, the result is pretty nice. Kodak has some excellent algorithms under patent protection right now.

Low cost compression on-the-fly would really help. I will look into Cineform. Our SPIHT doesn't apply well here - very processing sensitive and not much of an improvement over JPEG2K for most applications.

For image lattitude, we have some with dual slope capability (resets overly bright areas part way through the integration) but the result is nice for security - doesn't really extend the dynamic range - it looks strange.

Large pixels cost more - silicon costs are mainly real estate and yield. Big chips are tough on both. Of course you are right - the coupling is better. When I use my Canon lenses on my c mount cameras, the FOV is much narrower due to the smaller sensor. Are there adapters with relay lenses?

Let's see where we can go at the cost/performance level that most people can reach before hitting 8Mpix.
Steve

what you need is a ground glass system like I just made.. that way you can shoot with lowcost sensors that are small/cheap but use the 35mm lens system that gives you the look of film from the DOF 35mm priovides

To Les
 

Thank you, Les; a lot :)

Say,

Don't know, but figured this might come in handy to some of you guys

http://video-equipment.globalspec.co...ecurity_camera

maybe?

<<<-- Originally posted by Steve Nordhauser : OK,
I asked for information. The problem is that not much comes free. There is a threshold where the sensor costs are not the most significant system cost. PCs are getting cheaper but when you start pushing a bus and mass storage above about 60-80MB/sec, there is a cost step.
--

I assume this is where a second/third HDD is added, the via boards do have some virtual raid features.

Maybe the answer is to prioritise the requests on cost and performance and then resolution. Of course the trade off for people is wherever to have 3 1/2 720p chips or one 35mm 1080 chip.
---
For image lattitude, we have some with dual slope capability (resets overly bright areas part way through the integration) but the result is nice for security - doesn't really extend the dynamic range - it looks strange.
---

Yes, your right, it has to be done progressively to look right (no big jumps).

---
Are there adapters with relay lenses?
---

Yes there are relay, and condensor/feild lenses, possibly using special directional projection screens, much is being done in other threads recently, I made a bit of a summary of the little that I know over in the Viper thread.

---
Let's see where we can go at the cost/performance level that most people can reach before hitting 8Mpix.
Steve -->>>

Yes, my last post really breaks down to 3 levels 720 1080 and SHD, 1/2, 2/3, 35mm, each progressively more expensive. I have posted a lot of links in the viper thread. By the way what is the maximun average compression ratio for lossless?

About the 2/3rd inch sensors, I would settle for 1/2inch and use a 35mm/MF lense adaptor, but I included it because I think some people would prefer to use straight lense systems.

Thanks for the advice Steve, I had been curiouse about the price range of the splitter prism.

Laurence, really good find.

http://www.dvinfo.net/conf/showthread.php?s=&postid=183804#post183804

<<<-- Originally posted by Valentin Wegerth : well at least a "2 plane imaging system" could be done easily and wouldn't be expensive. only thing you'd need besides 2 chips is the body of a slr camera. any slr cam has 2 image planes. if you take a picture the mirror that normally rests in a 45° angle is removed and thus the image (or focal) plane is at the very point the 35mm photo-film usually is transported. if the mirror rests in its 45° postition however the image is reflected to the viewfinder (2nd image plane).
i hope you get the point... now if you'd just remove half of the mirror, half of the picture would reach the 1st plane where our 1st cmos chip would be placed and the other half of the image would be sent in a 45° angle to the 2nd image plane - given the viewfinder removed . and could be captured there with our 2nd cmos chip. once again we use the merits of 35mm DOF (see my last posting). now if we put both images together there should be no noticable 'cut' or breach because either the image - actually the light - passing the lens is sent to the first cmos or reflected in a 45° angle to the second cmos.

any comments on my thoughts appreciated ;) -->>>

I said:

<<<-- Originally posted by Wayne Morellini : Good idea, you could simply keep going with this to get as many mirrors as you want. The edge of the mirror would have to be machined smooth and tappered to the front to reduce distortion (but will it cause diffraction?? patterns). It would have to be sealed air tight, it would be bad to clean.

This gives me an idea. If you had a half mirored surface (no backing) then half the light would be reflected to the veiwfinder and half to the film plane, going one step further the mirrors could be coated to only reflect an individual range of primary colour, using two mirrors, who needs a $1-3K 3 chip prism splitter? I'm going to bring this up in the Home made camera thread. -->>>

Now I have another idea, something like a lenticular array could be designed to split the individual direction the image primaries are projected to be read by three chips or three areas of one ;) A bit of single direction image compression would do the trick. This is based on an idea I have had a long time ago (as well as a projection idea): a lenticular array laid over a single sensor could take all the light per pixel area, mix it and split the primaries to 3 ajioning pixels (acting like little prisms). What you get is a completely acurate (less any abreviations) colour, each primary is sent 100% to it's own pixel (no major filter loses), all on a single chip, at very cheap price. The other advantages is that you get near 100% pixel area coverage, not 70% max, like in cmos, so you reduce the fly screen and bayer motion induced luma/chroma artifacts, and increase the used light (if you design it to miss the interpixel spacing. The other benefits of these screens is that they could be used as projection screens for the adaptors. I think I gave up on the idea after the foveon came out. So would this reduce the costs? When done right (with a couple of other adjustments) you could deliver all the transmittable light from a MF lesne right down to 1/2in chip. I have other ideas I am wanting to work on commercially aswell.

thanks

Wayne.

Say guys. How would a 1280 X 1024 chip do us as long as it ran 24p I guess we'd crop the image to a 720p image right? Got a possible lead on a camera that will do this.

I'm waiting to see what comes out in Bayer masked cameras. I'd like to see something that has 2048 across , and maybe has some data reduction on chip if that's possible. I don't like the complexity of 3 chip cameras, I think the Bayer cameras will eventually make the 3 chip guys fall into the video tube category.
I actually have three ccd cameras that are Bayer and are 3500 by 2300 pixels, but they are about 3 frames a second max. I'm converting two of them to mono ccd's for some new film scanning equipment for my company.
When can someone post a still Bayer pic from one of the current cameras, at 10 bits ? Even better, two images of the same scene so we can see the noise figure.
The CMOS cameras and sensors are cropping up like mushrooms lately, it's going to be interesting.
-Les

Laurence,
If it is a CMOS chip, it will probably have Region of Interest (ROI) capabilities built onto the chip. One of the advantages of CMOS is that the process is the same as IC manufacture which allows gates (registers, counters, comparators, A/D) to be put on the same chip as the sensor. This means that the window size and location (digital panning) can be dynamically changed.

In most parts, windowing in the Y direction will allow a faster frame rate. Some parts require the full line to be shifted, regardless of ROI.

This speedup can be used to your advantage, even if you want to go 24fps. I think it was Obin who suggested that to minimize the rolling shutter tilting artifacts, you want to readout as fast as possible. If you readout at 2x the required frame rate, you can toss every other frame. It still isn't global shutter, but it is cheap and available.

Interesting.

Yes, it is CMOS. Hmmmmmm

larry, maybe you want to check out my thread called 10bit 4:4:4 I am building a CMOS 1280x1024 camera right now..

Update:

got the K3 16mm camera today...hmm a well made unit I must say...not like a plastic toy... very heavy like a russian tank ;)
anyway i see a big issue ...if I remove the shutter then I remove the ability to see the image in the viewfinder....any ideas for this problem anyone?


when the shutter is open you see nothing in the viewfinder when it is closed it has a mirror on it that shows you the image in the viewfinder


I guess I could try and use a 2 way mirror but then how many f-stops would I throw out??

what mirror or beamsplitter should I look at if any?

http://www.thorlabs.com/SelectGuide2...d=beamsplitter

maybe the amount of light coming in with a beam splitter would be ok because 3ccd cams have beamsplitters right? so should I look at 50/50 splitters or?

Obin, I think you are really determined to use a real camera body for this, even though you will probably end up with using it as a metal box and nothing more ! Form follows function, and you are making function fit the form.... hehe.
Some of the best and highest quality photographs come from 8x10 cameras, and they are nothing more than a wood box with a lens screwed on the front. Thats all a camera is, really.
It's all good, as long as it works... go for it !

...if I remove the shutter then I remove the ability to see the image in the viewfinder....any ideas for this problem anyone?

I guess I could try and use a 2 way mirror but then how many f-stops would I throw out??

Obin,
Have a look at 16mm Bolex cameras. They employ a beamsplitter (60/40). Maybe you can find somewhere one as a spare part...Their shutter is in the focal plane of the film gate.

I would also second Les' point. Why do we have to look at a design based on film technology requirements?

why? because when you pick this thing up it feels like a camera not a toy and I don't want to shoot with a toy. It's a great lens and lens mount and has lots of room inside for the HD board camera to fit...Why is this wrong? Would you have me build a square box to shoot clients big budget spots with??

The Bolex uses a beam splitter...can you send me a link so I can see how this works? 60% to the film and 40% to the viewfinder?

what is a good optical beamsplitter ? anyone know about them?

Obin,
Have a look here for the Bolex manual:
http://www.evergreen.edu/media/pdf/Bolex_Manual.pdf

awesome...bolex uses a 20/80 beamsplitter...i will look for that

I have seen simular suggestions in the static adaptor link. We allready have a light loss, and now we have even more, it makes more sense to use an external monitor (or the magnetic monitor flipping technique, external mirror depending on camera) and correct in post, as whenever you do these things you get optical compromises.

Now Steve wanted to discuss the issues in this thread that was to list just camera and camera mods, so how is it going Steve (when you get back from work tommorrow).

Thanks

Wayne.

Lens mount
 

Where would I find a lens mount for a custom-built camera? Is there a supplier for components like this?

If I'm going to use 35mm SLR lenses, I could just buy a non-working camera body from EBay ...

A Ballows lense system was one suggestion (or make your own rail pack) there are links to others in the 35mm adapter thread.

hi everyone

I saw this ultra mini pc at The Vision Show East. It may not suit our needs, but the size is perfect...

http://www.fase.co.jp/IEEE1394en/PC-CUBE/

It is a slick computer, no doubt about it. The big failing for these applications is the lack of a good RAID interface. Under "IDE interface" they list one CF slot. The same box with the chipset that supports serial ATA RAIDs off of the southbridge (ICH5R) would be great. That keeps the data off of the PCI bus, where most of the limitations show up. That would be the ticket to a low cost, fast system.

sent them an email asking about SATA on the system...will let everyone know what happens...Steve it has gigabit...whynot pipe into that network port and record direct to disk on the PC-cube? the hard disk inside is not fast enough?

Hey Wayne,

Is he talking about the Summix camera? One thing I was curious about is wheather or not FCP HD could actually use the stream coming from whatever camera is being talked about here. It was my assumption that "DVCPRO HD" was a specific standard that the FCP would identify and therefore could interface with, as opposed to a generic 960 X 720 stream. I figured a generic stream would not register the same and thus would not be compatible, or am I wrong? If I'm wrong, then you could also use the same system for the 10 bit 4:2:2 1080p, but if I'm right, well, then no. Am I correct in assuming it's a specific codec that must be used? What's the status of the summix camera anyway?

For that matter, I'm getting kind of lost in all these different alternatives. Maybe you could make a quick summary of all the solid possibilities for me? (I know, I ask so much, but I'm just a little filmmaker in a big, scary world).

Thanks!

Laurence

As far as I know he was talking about the Sumix camera originally (I haven't read it lately). There are at least 2 cameras going in threads here (I think some individuals are also getting their own heads to make there own). If you look at the viper thread I address some of the alternatives, problem is it is in Sumix's hands and we don't really know what they are doing yet. One thing for certain they were going to sell a camera box, no comrpession, everything else has to be settled by the buyer. Maybe he knows something we don't, or he is talking about another camera.

I have personally said to Steve that maybe we should just settle for a compresed 400Mb/s steam (if available, simply giving up on higher than 720p RAW 4:4:4 streams. Maybe Firewire 800, or 1 Gigabit ethernet streams). I share your bewilderment.


Richard
Cute, I wonder how many people would confuse it with a power supply if I left it on a desk ;)

Here is one of the cases I was looking at:

http://www.eyo.com.au/details_G-3688BK.html

I also was thinking of using another cube case and using the drive bay to house the battery.

But case modders make their own (and guess what I want to do).

About the limits of transportable computer:

Steve N,

I have heard of software RAID, or Virtual Riad (VIA), but haven't had time to research them, how pratical are they for what we want to do with upto 4 drives (the consumer IDE limit).

About processing, you maybe able to get two future VIA processors at 2 Ghz running in parralell (but in main baords), at a lot cheaper and more portable, lower power cost than a twin Intel server soluton, would that be able to help the situation? Your site advocates Cameralink as an interface, but unless we have a cheap card for that, would a multiple link USB2.0/3.0, Firwire, or Ggabit Ethernet be a viable cheaper solution for users. There is talk of satuating the PCI bus, but many advanced internal PC bus architechtures go beyond that (I think VIA/s internal bus was 266MB then 1GB/s and I don't know where it went from there, and AGP, Intel's PCI-Express and AMDs greatly exceed this) so if these interfaces by pass the PCI bus it need not be a problem.

Thanks

Wayne.

Here is a link with tests regarding a RAID O array with 4 SATA drives (they claim a 200M/s Read - Write sustainable speeds):
http://www.barefeats.com/hard35.html

<<<-- why not pipe into that network port and record direct to disk on the PC-cube? the hard disk inside is not fast enough? -->>>

Yes, I think that's the basic problem. The SATA 150 standard will run at 150 MB/sec, but IIRC, most desktop drives won't handle above 50 MB/sec or so.

For uncompressed HD, we need between 80 and 120 MB/sec, depending on the bit depth and frame rate. We will also be limited by the sustained throughputs of Gigabit Ethernet (approx 115 MB/sec) and Firewire 800 (approx 80 MB/sec).

Valeriu's link shows that a relatively inexpensive 4-drive array can handle 184+ MB/sec on a sustained write, which should be plenty. (A 3-drive array might be enough.)

Wayne,
There are two reasons why we are proponents of camera link at the high end (we do USB 2.0 at the low end and GigE for medium speed). First, honestly, most of our applications are industrial and camera link is *the* standard in non-cinema connectivity. Second, it is very fast and extendable. Base camera link has 3 8 bit channels which can be used as two 12 bits. 64 bit frame grabbers are available with dual base or full interfaces which doubles that. Data rates are high - the standard clock goes to 66MHz (at 24 bits) but the newer cards can do 85MHz. This means that you could run the AltaSens chip - 12 bit, dual tap 75MHz over to a single frame grabber.

Cost isn't so bad - our basic bundle with a camera (32 bit frame grabber, 2m cable, power supply and cable and some basic GUI software) is $500. The 64 bit/66MHz frame grabbers start at $1200 but will be dropping soon.

I've been talking to Xilinx (FPGA manufacturer with an HD-SDI demo board) about HD-SDI based on suggestions in this group but the spec maxes out at about 75MHz of data - you could do 1920x1080x30fps but not the 60fps the AltaSens is rated at. The design uses a Virtex II part - pretty expensive. This would be more cost than camera link. I'm not sure how 12 bit support works - the apps guy was talking 10 bit.

You can also get a cardbus camera link interface, but then you need to solve the disk recording rate if you go to a portable. There is also a PC/104+ frame grabber available to make a small card stack portable recorder.

GigE has its benefits - the best being 100m from camera to PC. Some of the latest chipsets from Intel put a gigabit interface off the southbridge - no bus bandwidth as you suggested. Still only 800Mbps continuous. Again not up to 2MPix x 12 bit x 60 fps, even with data packing. Not even 8 bit. Maybe 12 bit packed x 30fps but that will be close.

<<<-- Originally posted by Steve Nordhauser : Wayne,
First, honestly, most of our applications are industrial and camera link is *the* standard in non-cinema connectivity. Second, it is very fast and extendable.

Cost isn't so bad - our basic bundle with a camera (32 bit frame grabber, 2m cable, power supply and cable and some basic GUI software) is $500. The 64 bit/66MHz frame grabbers start at $1200 but will be dropping soon.
-->>>

Yes that is why I like the camera link interface, it is a good alternative standard to the very expensive SDIHD, but how expensive is it really (I know that it is limited and that is why I don't personally like it). If we can start up an alternative standard in interfacing (anybody that wants to interface to standard TV industry interfacing can pay the extra fo it, but we don't need it to do work). Most of my suggestions of late piont to an alternative work flow to the industry standard.

Now that $500 price cameralink (which I take is if you buy a camera as well), is still $500 more than USB2.0 ($100 I would not even care). So what about my suggestion of multilinking USB2.0, Gigethernet (but wht main board will; do that), SATA RAID/virtual; RAID/Software RAID, or cameralink board on PCI-Express, or AGP, is that possible? I personally have seen a reference to some new consumer HD serial interface that does around 5Gbit/s. Then there is the 10 Gigabit ethernet standard, is that comming to a motherboard chipset anytime soon? Sorry to harp on about it, just thinking of the whole market, top to bottom, would probably be good for industrial applications aswell. Look forward to seeing example footage from your cameras here in the near future.

I personally think that single chip true 720 and acceptable 1080 is good, which works along the data rates you mentioned. When these things get blown up big screen true is better unless you are going above 1080. If there is a propper cheap interface/software workflow solution I think there is market for single and 3chip boxs.

If you go to the link at the begining of this thread tot he Russian network camera, he has some cheap opensource FPGA compression designs for gigbit erthenet in his cameras.

Sorry for this bulk of writing Steve, it is late.

Thanks

Wayne.

<<<-- Originally posted by Steve Nordhauser :

64 bit frame grabbers are available with dual base or full interfaces which doubles that. Data rates are high - the standard clock goes to 66MHz (at 24 bits) but the newer cards can do 85MHz. This means that you could run the AltaSens chip - 12 bit, dual tap 75MHz over to a single frame grabber.

Cost isn't so bad - our basic bundle with a camera (32 bit frame grabber, 2m cable, power supply and cable and some basic GUI software) is $500. The 64 bit/66MHz frame grabbers start at $1200 but will be dropping soon. -->>>

OK, just for clarification -- will one of your existing frame grabbers handle the full resolution/full frame rate of the AltaSens 3560 -- 60 fps? Which one and how much does it cost?

Also ... you mentioned that you have new cameras coming out soon. Can I assume that one of them will be based on the AltaSens 3560 chip? Any idea about price range?

Also also ... I notice that some of your frame grabbers come with an SDK. How is the SDK licensed -- can we develop GPL'd software with it?

Thanks!

Steve I need you to call me asap I have your camera and need help!
thanks!

<<<-- Originally posted by Rob Scott :
Also ... you mentioned that you have new cameras coming out soon. Can I assume that one of them will be based on the AltaSens 3560 chip? Any idea about price range?

Also also ... I notice that some of your frame grabbers come with an SDK. How is the SDK licensed -- can we develop GPL'd software with it?

Thanks! -->>>

Yery good pickup Rob. You know that those ITX boards have PCI, I am wondering how many pixels hor/vert to get an effective resolution 1080 picture, and can we squeeze it through one cameralink PCI frame grabber card?

I also would love to see what type of cameras they release shortly in their HD range.

Does anybody have an updated list of the best lowcost sensors, I would like to veiw them, their has to be a stand out value sensor out there somewhere? Wonder what Foveon is doing?

Wayne, seems to me that Micron is doing the best lowcost chips..I have the Silicon Imaging camera and am going to start some tests! foveon is 4fps I think and the fillfactory chips suck ass in image quality the rockwell chips are great but cost alot