Idea for TriColour videos

[Stefano]

Hi, I have had this idea for a while, I'm sure this isn't anything new, but I wanted to share it.

 

Using a camera with a high frame rate (at least 120 fps) and a rotating filter wheel, it could be possible, at least theoretically, to make a TriColour video. For example, having 4 filters for three wavelength bands (one doubled, ideally for the band at which the camera is least sensitive), at 120 fps would allow to take images at three different wavelengths 30 times per second, and with some digital processing have a color video at 30 fps.

 

For UV especially the camera would better be monochrome for sensitivity reasons. If the blue channel is chosen to be UVB, taking a UVB photo at 1/120 s exposure time could maybe be possible using a fast lens and high ISO. See Lukas` post for UVB video. It's doable with a monochrome camera and I strongly doubt you can do it with a color camera, unless you are recording the Sun.

 

The resulting video would probably suffer from color fringing for fast-moving objects, and also there's the problem of different sensitivity to different wavelengths (one solution would be ND filters for the "brightest" ones).

 

I don't have such a camera and I won't try this anytime soon, if I ever will. Has anyone tried this or is going to?

[Andy Perrin]

That is how some early color TV systems worked I believe?

[dabateman]

I think you could only do this using three sensors,  with three filters. Like the Panasonic 3ccd video cameras.

Using a Raspberry pi,  that might be possible and not too extremely expensive. 

Rapidly rotating your filters will cause problems. And exposure will need to be different for each filter change.

[Nate]

I have a couple astro cameras, ASI462 and a ASI178MC(uv/ir cut removed) that when I use something like sharpcap or firecapture, which are free capture programs to run  the cameras and make the ROI(region of interest) smaller, I can get up to 240fps. Playback at 30fps makes for some pretty slomo. 

 

I haven't progressed enough to get a filterwheel, but someone might be able to try it with those cameras.

[lukaszgryglicki]

I think that camera won't be able to adjust to brightness differences at that crazy rate.... if I put a filter on my mono camera while looking through EVF the image becomes pitch black and brightens in a second or two; and if I remove filter then it becomes whiteout for a second or two, doing this with 120 FPS is impossible IMHO, uneless I miss something or misunderstood.

 

EDIT: with just 3 "similar" brightness filters (like R, G, B) it seems doable, but with UV, Vis, IR it would be like: UV=5 stops darker (or more), Vis=reference, IR=about the same.

 

 

[Stefano]

Yes, that's why I suggested ND filters as a possible solution.

[Stefano]

On 3/28/2023 at 4:08 AM, dabateman said:

I think you could only do this using three sensors,  with three filters. Like the Panasonic 3ccd video cameras.

Using a Raspberry pi,  that might be possible and not too extremely expensive. 

Rapidly rotating your filters will cause problems. And exposure will need to be different for each filter change.

I think the hard part would be to split the image into three channels. Could you use dichroic mirrors?

 

This would allow a 1/30 s exposure for 30 fps video, can a Raspberry Pi do that in UVB?

[dabateman]

2 hours ago, Stefano said:

I think the hard part would be to split the image into three channels. Could you use dichroic mirrors?

 

This would allow a 1/30 s exposure for 30 fps video, can a Raspberry Pi do that in UVB?

Yes dichoic would be the way to go, as prisms would be hard to isolated the wavelengths. 

I don't know in sunlight,  never tested. But controlled light, yes.

This is the Panasonic 3ccd diagrams:

[lukaszgryglicki]

And in this case where UV goes and where IR goes? This isn't that simple as UV in B and IR in R right? Or is it?

 

[Stefano]

If you use that prism, I guess so. If you want to build such a camera for UV, you would choose appropriate dichroic mirrors.

[Stefano]

Also there's another issue: the lens needs to be physically far from the sensors, so wide angle lenses might not work. Probably a retrofocus design is needed for such lenses.

[KaJashey]

Just chiming in that dicrotic prisms were indeed the way to do 3CCD cameras. They were always smaller sensor cameras and I think they died out doing 1080 video though most were 480 DV. I might have a Sony TRV900 somewhere with the prism at it's heart.. 

 

There were color seperation cameras in the 30-50's that I believe used two sheet beam splitters (like DSLR mirrors) and color gels.. like this and this Required a lot of light.

 

They look a lot like the 3CCD dicrotic prisms I wonder if the angles are similar.

[OlDoinyo]

The beamsplitter idea being discussed here sounds like a slightly more complicated version of J.W. Wong's setup for IRG. Of course, an extra beamsplitter would need to be added for this to work, and such a setup is not conducive to the use of wide-angle lenses (as noted above.)  If one of the channels is UV one would need to be careful; probably the first reflected beam would have to be used, so that fraction is not forced to go through any unnecessary optics. Prisms might be especially problematic as they tend to attenuate or block UV. The beamsplitters would have to be bespoke angle-calibrated dichroics (which are different than ordinary perpendicular dichroics, although such things exist.)

 

The rotating-wheel idea has its own challenges, including precise synchronization of the wheel with the sensor scan cycle (most consumer cameras have no obvious way of achieving this) as well as the fact that there will be a portion of the rotation cycle wherein one filter sector is not cleanly in the optical path (will the sensor cycle have a long enough downtime between frames to accommodate this?) One supposes that some channels might need to be attenuated with ND filtration to match the much lower sensitivity in UV. But this way would enable the use of wide-angle optics, and might actually be the more promising if UV were involved.

[dabateman]

I actually have a Panasonic 3ccd camcorder.  My son dropped it when he was little,  but I have kept it arround. I hadn't thought to take it apart.  It might be a fun project to see what the wavelengths transmitted are. I wonder if the blue channel is just a short pass and the red channel is a long pass. That might help. Mine is the PVGS200,  designed with 1/6 inch sensors. So the prism will be too small for my 1/2.3 inch pi sensors. 

[Andy Perrin]

When I tried using a stereo adapter (mirror-based) for taking a UV and IR photo simultaneously, I ran into issues with the huge exposure difference needed between the UV and IR. For still images, I just edited each side of the photo separately and combined them afterwards in photoshop, but it might be more complicated for video. I imagine you could adjust for this somehow by varying thickness or ND-filtering the IR side?

 

 

[lukaszgryglicki]

If something uses filters like short-pass, long-pass and ir/uv cut then I'm sure all of them will have out of band leaks (as most UV pass have IR leaks, opposite is not a problem usually like IR pass leaking in UV).

But if this is using prism then I imagine that splitting UV+B and shorter then Green visible an dfinally R+IR should work (by the prism definition) ... ?

 

[Andy Perrin]

Yeah, leakage would not be an issue here because the light is already split.

[Stefano]

Despite all, this actually sounds doable. You need to align everything properly, synchronise the sensors, adjust the exposures for each one, and combine the data later, but it can be done provided the mirrors are available and you can record in UVB.

[lukaszgryglicki]

For UV-B I've found the *biggest* issue (for me) is filtering everything else out. I know no out of the box single filter that does it right. Once I have UV-B filtered out - I can shoot video directly.

 

I would rather think about using tri-tri pod with 3 idependent cameras (close to each other) one for UV, one for Vis and one for IR. Having 3 cameras will also make 3D rendering possible (choose any 2 of them for stereoscopic view).

 

Then I would combine frames in post-process.