Viewfinder options (DIY-HD-CAMS)

[Oscar Spierenburg]

I thought I'd start a new thread on this subject. The best solution for home-brew HD cameras is of course a real time stream directly from the sensor to a monitor.
But because most projects around here (on DVi) are still in the stages of getting the primary things like capturing and compression etc. to work, I'd like to discuss alternatives to get things started. If we have alternatives, we can focus on a final solution when the camera itself works.

I'm working on the Elphel project and made a start to solve the preview problem by making a real optical viewfinder on the camera.
It seems to work, but a lot of adjusting still needs to be done. Here is an image of the current stage: http://www.dvinfo.net/conf/attachmen...0&d=1200170306

I use a beam splitter right in front of the sensor and direct it though a mirror a 'flip' tube to get the image right way up to the eye piece. Almost all parts are taken from a super8 film camera.


I was also thinking about using a battery powered mini security camera at the end of the beam splitter and show the video output on an LCD monitor or an electronic viewfinder from a video camera. This may sound strange, but you'd have the opportunity to rotate the monitor when you're using a 35mm adapter. Of curse it's not suitable for focusing because of the resolution.

Anyone got other ideas?

[Samuel Hinterlang]

I like your direction here. You will see whats coming in the lens. There seems to me like it has its downsides though. This has to cut a couple stops of light out, right? You might also be seeing some of what the sensor is not getting.

The few times I went down this road I could only think that is you could get a studio camera type view finder to work off the back end of the camera so you could see exactly what the sensor was getting or even at the worst case to have a composite rca out to hook up a cheaper end head mount display. But, these would be totally dependent on the kind of outputs on the camera itself.

Really you just need some way to get a live image off the camera (vga, hdmi, composite, cameralink, whatever).

[Oscar Spierenburg]

Yes, live image off the camera is the best and final solution, but in the mean time we should work something out so we can actually use the camera while testing etc.
I don't know about the Sumix camera, but I think they don't have any preview or other direct stream either.

I've done some adjustments on the optical viewfinder. It's hard to get everything right. Does anyone know how to built an extension tube for a viewfinder?

You don't lose much light with the beam splitter, because it reflects only 30% or less light to the viewfinder.

[Sebastian Pichelhofer]

I do like the idea of an optical viewfinder. Makes filming feel more like the good old celluloid days or like working with a (digital) ARRI camera like the D-20 BUT:

-) as mentioned you loose light

-) you still need some kind of digital display or scope to get the exposure set properly

-) Viewfinder != Viewfinder, you need an optically well designed and big!! viewfinder to use it for focusing and to get a good picture that is a pleasure viewing. The 8mm Viewfinder seems to be pretty small and the viewing angle can't be too big (I remember trying to look through the eyepiece of some old 16mm camera, it felt like trying to see something through the ocular of a cheap, tiny microscope )

-) The viewfinder adds another way for light entering the camera case and therefore getting to the sensor. The eyepiece must be completely covered either by your eye or something else when not in use.

[Oscar Spierenburg]

That's true. We can only use this type of viewfinder for framing the shot.
On the other hand, maybe it could be enough for the Elphel project, because there is a new feature in development that indicates 'focus-quality' (in the control panel software) If that could be displayed, and information about the exposure levels, it could be enough to get started with actual film making.

I've received one of those IR pinhole cameras (like this one and will put that in place in stead of the viewfinder eyepiece. Attached to an lcd display or a viewfinder from a camcorder, we'll have a low resolution, but bright image of what we're filming.
I've tested the Elphel camera with the beam split and it doesn't seem to loose much light.

[Sebastian Pichelhofer]

Originally Posted by Oscar Spier

there is a new feature in development that indicates 'focus-quality' (in the control panel software) If that could be displayed

As you said this would still require some kind of display. So why not use the display for the actual video as well :)

But the focus-quality can be used as overlay just like a histogram to judge exposure.

[Oscar Spierenburg]

This is an old thread (thanks to the moderator for moving everything!), but here's a small update on this subject:

I recently bought a Pan-Cinor lens on a flee market for 25 euros. The extension tube is missing, but still, beautiful lens: http://www.bolexcollector.com/images...s/60som_09.jpg
This originally attached to a Bolex 16mm camera (c-mount), so it would only be of use on a 2/3" sensor... but who knows;-) I couldn't resist buying it. It's also good to study it, in case we decide to have a custom relay lens made to use a 35mm adapter. An integrated optical viewfinder is not a bad idea. One clever thing is that the beamsplitter is behind the iris so it doesn't effect the viewfinder brightness. (same as in 8mm cameras)

[Bob Hart]

Oscar.

If you go for direct relay, a quick and easy is a 4+ approx achromat and either Voigtlander for Nikon 40mm f2 or Nikon 45mm f2.8.

The P+S monitor for the SI2K can be sourced without the breakout box as an in-car computer screen from "CarTFT" in .de. I think it is called a CTR840 or something similar ( can't remember right now).

If you want to follow it up (USB Touch Screen, VGA and composite, component and AV sound in) let me know. I am off internet for about a week due to provider problems.

[Biel Bestue]

maybe the solution is not hardware-like maybe the solution could be software-like, is there a way to show the camera operator what is in focus by emphasysing what has high frequency detail like the si2k?

another solution that would be hardware needy, would be to get the computer establish where the focus is, kinda like boujou and matchmoving software does, but sort of realtime... ok, maybe this is for the future... :D

But the si2k shouldn't be that dificult...

to tint those in focus pixels (of hight optical frequency) of inverse color (or complementary) and bring brightness to higer value than the surrounding pixels

i guess that system wouldn't work with low freq. surfaces that are in focus, but it would work most of the times

i don't know how to bring this ideas to life at the moment, but i do know "S.T.A.L.K.E.R." videogame used some kind of pixel shader trick to determine where the render had aliassing in order to blur the image, maybe some code form that could be used to determine where is that high freq detail, this could be found in one of the nvidia papers in their developer site, now i just can't find it

this also means one could work with little screens that aren't even fullHD, maybe something in the 720p or the 480p range

what do you think?

[Oscar Spierenburg]

:-) That's exactly what Sebastian and Andrey are working on! Andrey calls it 'focus quality'.. Sebastian calls it 'focus Peaking'.

[Andrey Filippov]

Yes, there is some FPGA code to determine focus quality (or amount of high frequency components). It uses the 2-d DCT coefficient that are calculated for each 8x8 pixel block as a part of JPEG compression. Additional code calculates weighted sum of coefficients squared for each 8x8 block, "weighted" here means multiplied by arbitrary set ob 64 values (can be loaded through PHP code). So the result is intended to be indication of high frequency components in each block.

Next part of the code can add the result to the image, effectively highlighting the boarders (filters can be set to detect any, only vertical or horizontal gradient, have different "cutoff frequency"). That combining of images is "destructive" (replaces original image/video), but it can be turned off and only the overall "focus quality" value used as described below.

And the last part of the FPGA code calculates sum of those per-block values over the specified rectangular window, the single per-frame value is available to the software. The demo PHP script just plots that value (in logarithmic scale).

[Biel Bestue]

now, what happens to low frequency infocus features? how can one get info from non contrasty situations? maybe somthing relating to pixel flow to determine the 3d in the scene when in movement in case of movement?

[Sebastian Pichelhofer]

Originally Posted by Biel Bestue

now, what happens to low frequency infocus features? how can one get info from non contrasty situations? maybe somthing relating to pixel flow to determine the 3d in the scene when in movement in case of movement?

It's pretty simple: if your eye cannot tell if something is sharp the algorithmn will fail as well.
I remember this picture of a desperate focus puller looking at an entire room with all walls covered with thin fine red fur. No sharp borderes, edges or constrasts :)



Back to the viewfinder topic:

So far we considered 2 possible solutions:

-) a tablet netbook
pros: big display with moderate resolution, built in battery pack,
cons: low performance (especially with intels built in onboard video cards), good enough for viewing 720p video but not fast enough for 1080p (<15fps), tablet netbooks with touchscreen are still rather rare, no analog video outputs, no HDMI outputs (not that I know of), display and computing hardware cannot be seperated

-) beagleboard with LCD
pros: HDMI output, analog video output
cons: even less video playback performance (just enough for 720p), requires external battery pack, requires external display and touchscreen controller, no built in ethernet (usb adapter is too slow)

Now here is a 3rd options worth considering:

-) an Nvidia ION ITX mainboard in a compact housing
high video playback performance due to geforce 9400 chipset (should be able to handle 1080p playback), more power than netbook as dual atom cpu available, power consumption of around 20-30W, dc-dc converters for powersupply are available, HDMI output, can be equiped with additional peripherials like HDDs, CF cards, etc.
Would also require an external display with touchscreen but that way the display can be mounted somewhere else as the CPU.

So the ITX solution would be similar to the beagleboard but with a fast Ethernet port (Gigabit on most products I found), which will be required as soon as Elphel 373 is released and higher overall performance.

[Dan Hudgins]

I think a good option for a viewfinder it to build a scan converter that uses SRAM LUT, two one for the sensor end and one of the output end, then you can output SVGA, NTSC, or PAL to use any monitor, and you can input any sensor resolution.

There would be two memory, you store to A while B is being output, AB flip when the sensor starts a few frame, B continues to loop while the sensor is busy or not working, that way the sensor can be in time laps mode or having a frame rate slower than the monitor, you can run the monitor at 120Hz and the sensor at 24Hz or slower.

VR glasses can be uses, and 1:1 zoom just requires re-programing the sensor address LUT, or swapping the sensor end LUT to change the area mapped into the AB scan converter buffers.

[Sebastian Pichelhofer]

Here is a more detailed plan of the viewfinder setup mentioned above:

Parts that we need:
*) Zotac ION ITX D Mainboard with dual Atom CPU and Nvidia 9400 GPU, passive cooling, ITX form factor (17x17cm)
ION ITX A Series

*) RAM Brick DDR2-800 (2GB sounds reasonable)

*) PCIe Mini Card SSD or SATA SSD (to keep the system as rugged as possible try to avoid any spinning parts like HDDs)

*) picoPSU, tiny, fanless, takes 12V input: picoPSU-120 12V DC-DC ATX power supply

*) Case either an off the shelf one or a custom design that can be mounted on the cameras rods support or at the back of the LCD.

*) 8.9" 1280x768 LCD kit
G&P Optoelectronics GmbH & Co. KG - 8,9" LCD TFT mit HDMI / DVI Controller (1280x768p) 10-02-25

*) + Resistive/capacitive touchscreen (8.9", USB preferable)

*) + Enclosure + Sunshade

*) Cables



This setup would look similar to Eric Wu's sumix camera setup (see attached image) powered entirely from battery pack.

Just that the Zotac ION ITX is much smaller than the computer case in the image -> 17x17cm, in comparison the LCD is 8.9" ~ 22cm in diagonal so it is around 18cm in width which is just around the same size the mainboard + enclosure would have.