Continouus shooting with 3D stereo possible? - page 2 - General Help and Assistance on using CHDK stable releases - CHDK Forum supplierdeeply

Continouus shooting with 3D stereo possible?

  • 26 Replies
  • 6274 Views
Re: Continouus shooting with 3D stereo possible?
« Reply #10 on: 04 / October / 2016, 19:01:23 »
Advertisements
If you search this forum for the term "precision sync" you will find may posts reporting actual results with real test data.  That data shows that two cameras that receive the same signal to shoot a single shot will frequently take hundreds of milliseconds different times to actually take the shot. 
Your data tells about synchronizing that first shot in continuous mode. Once you get that synchronized, I still say 2nd, 3rd, 4th and 5th are well synchronized. I may not be correct, but here is my reasoning:

Getting first shot in sync is difficult because of different USB cable lengths, times to adjust lenses and what not, and lots of microprocessor work prior the first shot.

So only syncing that first shot is tricky. CHDK uses tricks like triggering action after the release of shutter that is to give cameras enough time to prepare while the button is held down.

After that there's just a fast burst of taking another four shots with the same parameters and same amount of work for both microprocessors.

Re: Continouus shooting with 3D stereo possible?
« Reply #11 on: 04 / October / 2016, 19:20:57 »
After that there's just a fast burst of taking another four shots with the same parameters and same amount of work for both microprocessors.
I'm going to assume you have not done a lot of work with "real time" processing and microprocessors?  Canon cameras have many tasks all running independently under loose control of the operating system. Of course only one task actually is running at any given moment.  Some testing I did years ago showed something like 30 different tasks all active and running - albeit not always at the same time or intervals.   So the exact sequence of what runs and for how long is constantly changing and you thus can't depend on the timing not changing from shot to shot.

However, if you feel strongly enough,  please feel free to test this out.  You only need one camera and an oscilloscope. Just run a continuous shot sequence multiple times while triggering the scope on the first shot and observing the beam position on each subsequent shot.  I believe you will find a lot of variation in the actual shot positions.  Not enough for a photographer to notice, but more than enough to mess up stereoscopic images.

EDIT : for reference,  the tick timer on Canon P&S cameras runs on a 10 mSec interval.  But delays greater that 1 mSec between shots greatly affect the quality of stereoscopic images.
« Last Edit: 04 / October / 2016, 21:04:46 by waterwingz »
Ported :   A1200    SD940   G10    Powershot N    G16

*

Offline reyalp

  • ******
  • 12589
Re: Continouus shooting with 3D stereo possible?
« Reply #12 on: 04 / October / 2016, 22:23:45 »
Your data tells about synchronizing that first shot in continuous mode. Once you get that synchronized, I still say 2nd, 3rd, 4th and 5th are well synchronized. I may not be correct, but here is my reasoning:
This is not correct. The is substantial shot to shot variation in continuous shooting. The camera and SD card behavior aren't deterministic, and the precision of anything done from script is +/-10ms at best.

Looking actual data from scripts that shoot in continuous mode, I see a typical variation of ~30ms but there are outliers. It certainly doesn't meet the single ~1ms level desired for stereo.
Don't forget what the H stands for.

Re: Continouus shooting with 3D stereo possible?
« Reply #13 on: 05 / October / 2016, 02:11:19 »
Looking actual data from scripts that shoot in continuous mode, I see a typical variation of ~30ms
Thanks for showing my guess was incorrect.

30 ms sounds like a substantial number. How is any external timing circuit going to be able to correct on that I don't know.

If your figures are correct and your test was not faulty, then continuous shooting for 3D is probably out of the question. Can you tell me more how you obtained that figure? Did you read the times from file timestamps (if this is possible) or from let's say oscilloscope picture pairs? What was the exposure time? Did you experiment with "shoot -nowait" parameter as well?

And why is it your goal to sync within a millisecond when human eye can only detect 24 frames per second? Doesn't this 24fps resolution mean that human eye won't notice a difference for 1000/24=41 ms gaps?


*

Offline reyalp

  • ******
  • 12589
Re: Continouus shooting with 3D stereo possible?
« Reply #14 on: 05 / October / 2016, 03:23:35 »
30 ms sounds like a substantial number. How is any external timing circuit going to be able to correct on that I don't know.

If your figures are correct and your test was not faulty, then continuous shooting for 3D is probably out of the question
I don't think so. It just needs external sync.

In CHDK, the "USB remote hook" is able to stop execution very shortly before an exposure starts. The variation is mostly in the other parts of the shooting processes, so on many cameras you can get very good sync using the remote.

To use it in continuous mode, all that should be needed is to have the shutter held down, the remote code configured to wait in the hook until it receives a pulse, and something that sends pluses at a rate somewhat lower than the maximum continuous rate. E.g. if the cameras can typically do 1 fps, set the controller to send a pulse every 1.2 seconds. It might get out of sync if SD card has a big lag, but for the kind of use suggested by the OP I'd expect it to work most the time. A more sophisticated system could use some kind of feedback (e.g. LED + photo diode) to let the controller know when each camera was ready.

Quote
Can you tell me more how you obtained that figure? Did you read the times from file timestamps (if this is possible) or from let's say oscilloscope picture pairs?
I wrote a script which uses the hook mentioned above to implement an intervalometer in continuous mode. It logs timestamps for various parts of the process. The timestamps only have ~10ms precision, but given that the variation can be significantly larger than 10ms, I'm confident it's real.

Aside from that, I've been working on CHDK code for last ~8 years, so I know there are very good reasons to expect it to vary by that kind of amount.
Quote
And why is it your goal to sync within a millisecond when human eye can only detect 24 frames per second? Doesn't this 24fps resolution mean that human eye won't notice a difference for 1000/24=41 ms gaps?
The 24 fps thing is probably an oversimplification, but we're talking about still images here, so it's not relevant. What matters is the movement between frames. A person walking at 1 meter/sec would move 4.1 cm in that time.
Don't forget what the H stands for.

Re: Continouus shooting with 3D stereo possible?
« Reply #15 on: 05 / October / 2016, 12:42:08 »
- What about the synchronization of the the 2nd, 3rd, 5th photo - is it as good as the first photo when I trigger the USB-shutter?
It is as good.
Without external synchronization, two scripts running separate cameras will get many tens if not hundreds of milliseconds out of sync by the first shot, and diverge more from there.

You will be surprised about the results of an experiment that I got. The results will show that I was almost correct in my guesses what happens with consecutive shots in continuous mode. They don't "diverge more from there" as you said. In fact, they diverge and converge in random fashion.
I know this is a minor point in this discussion. Major point is how to get that first shot synchronized. Still, have a look at this tables. There is couple of tests that I made.





Picture says it all but here is some comments just in case:

Test A.
Since I haven't an oscilloscope I went to www.estopwatch.net, photographed the computer screen in continuous mode, and wrote down the measurements. Measurements show times when photos are taken.

Test B.
Interval shows time that elapsed between two photos.

Tests C-G show that the user is using continuous mode for the first time in his life. I have yet to learn here how to hold the button to make a proper continuous burst. There are some interesting intervals here so I'm leaving this results printed.
If you study intervals closely, you see a pattern here. It's about numbers 180 and 120 that appear randomly no matter what settings are used.

Test H
This is the test to prove your statement wrong. The pictures don't diverge more. In fact, they sometimes converge. In this example you get a couple of good anaglyphs (marked green) and a couple of bad ones (marked red).

Test H right camera
The ideal test would be if I had two synchronized cameras for this. I actually haven't and have a 5 hour delay with the same camera.
Doesn't matter, to compare the intervals this 5 hours are subtracted on the left camera so I got two cameras synchronized in Excel.

Now all there is left is to interpret the results. Cameras obviously are in sync in 1st shot, then they get out of sync later, because of this unusual interval (marked yellow), they stay out of sync until another unusual interval appears, then they magically fall back in sync.

Re: Continouus shooting with 3D stereo possible?
« Reply #16 on: 05 / October / 2016, 14:07:04 »
Major point is how to get that first shot synchronized.
Getting the first shot sync'd is easy.  The USB remote code in CHDK will get you within 1 mSec between two cameras almost every time.

Quote
Since I haven't an oscilloscope I went to www.estopwatch.net, photographed the computer screen in continuous mode, and wrote down the measurements.
There are several posts in this forum pointing out the issues with "estopwatch" for estimating sync.  It's just too slow (and the LCD refresh rates affect things as well) to produce sub-millisecond precision necessary to evaluate sync quality.

What the rest of your tests seem to show is random intervals between shots.  And the variability in that randomness is more than enough to make stereoscopic images of anything moving basically unusable.

I'll rig up a microcontroller and CHDK script tonight and see if I can produce more usable results.
Ported :   A1200    SD940   G10    Powershot N    G16

*

Offline reyalp

  • ******
  • 12589
Re: Continouus shooting with 3D stereo possible?
« Reply #17 on: 05 / October / 2016, 16:54:13 »
You will be surprised about the results of an experiment that I got. The results will show that I was almost correct in my guesses what happens with consecutive shots in continuous mode. They don't "diverge more from there" as you said. In fact, they diverge and converge in random fashion.
That was a poor choice of wording on my part. The bottom line is the more shots you take, the bigger the chance that some will be badly out of sync.

The real behavior I've observed is a smallish (few tens of ms) random variation and occasional long outliers (a few hundred ms or more). If you hit a long outlier on one cameras, everything that follows will be badly out of sync.

The long outliers are probably mostly related to SD card (they have their own firmware and CPU and do unpredictable things), although it isn't entirely clear all are. The first shot after boot also seems to slow and variable quite often.

In any case, the few tens of ms is likely to be fatal for stereo in many applications.
Quote
Test A.
Since I haven't an oscilloscope I went to www.estopwatch.net, photographed the computer screen in continuous mode, and wrote down the measurements.
Measurements show times when photos are taken.
As Waterwingz notes, your precision is limited by your display refresh rate. That renders everything that follows pretty much irrelevant.
« Last Edit: 06 / October / 2016, 15:49:53 by reyalp »
Don't forget what the H stands for.


Re: Continouus shooting with 3D stereo possible?
« Reply #18 on: 05 / October / 2016, 18:23:56 »
Uff, I see, that was some bad resolution Excel table indeed. I'm attaching a ball to a string now and nailing it to the ceiling. Tomorrow after I swing the thing I'll bring exact picture pairs.

Re: Continouus shooting with 3D stereo possible?
« Reply #19 on: 06 / October / 2016, 14:53:39 »

 

Related Topics