Tags:
  1. igoddard

    igoddard Active Member

    Here's an exercise in trying to simulate / illustrate a few of the shapes (on right) that rotate in the background (on left), which we've called 'bands'...

    [​IMG]

    One of the rotating shapes, the most prominent, appears in the upper left corner and seems to be a portion of a large ovoid shape. The other shapes are very amorphous and virtually impossible to illustrate, but we can still see them. There's a distinct impression of a square shape in the center, which I've partially simulated. The motivation for this is to try to quantify the rotating 'bands' as the only thing that bugs me about this evidence is the "Can you see it?" aspect of it. It's really, imo, a proof of Mick's initial hypothesis.
     
    • Like Like x 2
  2. igoddard

    igoddard Active Member

    A little more detail added to the patterns and then run for a loop with only the simulated patterns...

    [​IMG]

    It's hardly perfect, but I hope it conveys the 'gist' of the background pattern rotations. If you watch the original image on the left, you can see by peripheral-vision reference the right frame and thereby see what we're talking about in the original.
     
    • Winner Winner x 1
  3. Very nice.

    Someone suggested regarding Gimbal: "The noise in the video that looks similar to rotation is most likely a compresion artifact which recalculates every time the object moves or changes shape, because the speed of the object also changes so new keyframes for the codec that compressed the video will recalculate the image. You can try this yourself if you want, this is probably h.264."

    I created a quick and silly animation to hunt for any artificating that looked close to the suggestion.

    The first render to h.264 I used a bitrate of 1500 kb/s, then I took that mp4 and re-encoded it to 750 kb/s, then took that and re-encoded it to 325 kb/s and so on down to 175 kb/s (the lowest h.264 setting Adobe Media Encoder allows). Then for good measure, I re-encoded the 175 again at 175 kb/s.

    175 kb/s mp4 converted to an animated gif for forum (mp4 is attached):

    gimbal_mockup_175kbps.

    While I don't really see any rotating compression artifacts on the h.264 mp4, I did just notice some artificating (below the object and above the clouds when the rotation changes) in the animated gif.

    While not likely to the degree seen in the 12.4 Mb/s original, but since there is no chain of command and we have no idea how many times or in what ways this video has been encoded and re-encoded, is it possible the banding rotation is caused by compression artifacts?
     

    Attached Files:

    • Like Like x 1
  4. Mick West

    Mick West Administrator Staff Member

    No, that does not make any sense. An object rotating in the middle of the screen is not going to cause shapes to move in the corners of the image. Compression algorithms deal with local changes - i.e. changes within a region. The motion of the clouds is vastly more significant than the rotation of the object.

    Nice mockup!
     
  5. jarlrmai

    jarlrmai Member

    Yeah you should add some grain/banding to the the sky above the clouds in your mockup and retry the compression testing, it should show the true effect of compression artifacts as they handle noise poorly. Someone claiming the above would need to provide evidence as as Mick says compression normally affects blocks of the video differently.
     
  6. Agent K

    Agent K Active Member

    I see the color banding above the clouds disappear when the object starts to rotate because the clouds are less hazy at that time.
     
  7. Thank you, sir.

    Yeah, video\image compression has been an enemy of mine for longer than I care to admit. And while I know enough, I know that I don't know enough to completely discount the possibility of some sort of compression artifacting without consulting people who might know more than I do.

    I mean, I have a glancing understanding of macroblocking which is probably enough to know that the rotation of the small thing likely wouldn't cause rotating artifacts across the entire frame.

    But, the ignorance of the vocal arrogant was still enough to cause a second guess.

    jarimai, I did have some fractal noise in there on the cloud layer, but for some reason I thought that might be cheating somehow so I removed it. I'll try it again.

    Agent K, I think you're right. I'll change up the cloud layer and see what happens.

    Ok, two more things. I thought I asked here, or maybe not. What's supposed to be causing these rotating bands?

    Maybe I read or watched, or simply just assumed, that it was the result of the rotating glass/plexiglass lens covering/housing over the camera.

    But then why aren't similar rotating bands as apparent at the end of the video when the object's illusory rotation is more severe?
     
  8. igoddard

    igoddard Active Member

    The analysis I posted suggests to me that the cause is somewhat complex. The 'bands' are more complex than could easily be attributed simply to rotation of one surface like the plexiglass covering the housing window. Note especially the ovoid-like shape in the upper left, its rotation is not smooth and is not perfectly aligned with the rotation of the other shapes. And there are 'bands' in a number of directions, not all of which are consistently visible during all phases of rotation. So the exact cause(s) may involve influences of more than one surface.

    My generalized causal description would be: we're seeing artifacts of mechanical rotations in the optical system. I don't think we need to parse the cause(s) much finer than that. The thermal signature rotates precisely with these optical rotations, and so its rotation is manifestly an artifact of those mechanical rotations.
     
  9. igoddard

    igoddard Active Member

    The 'bands' do rotate at the end, but the rotation at the end so much less it's not very noticeable. The biggest bump to the whole scene (and thus to the camera) occurs at the end, but not the greatest rotation. Measuring in my editing program, the rotation over the time frame I spiced out above is ~90˚, whereas the rotation at the final big-bump moment is just ~25˚.
     
  10. Agent K

    Agent K Active Member

    Some combination of internal reflections and possibly sensor nonuniformities that escaped nonuniformity correction.
     
    • Agree Agree x 2
  11. Mick West

    Mick West Administrator Staff Member

    Internal reflections seem the most plausible to me. This might be another interesting thing to try to simulate. However the light path of the camera system is more complex than your average camera, with much more possibilities of odd reflections. I wonder if shooting though a small periscope might provide some useful comparisons.
     
  12. igoddard

    igoddard Active Member

    Here's a degrees-of-rotation analysis. The banding in the background seems to rotate slightly less overall than the target, but that ovoid shape in the upper corner is pretty clearly seen to do the same 90˚ of rotation, albeit it rotates in a somewhat different fashion. I'd surmise that the slight difference is down to the fuzziness of the bands, or in other words, I'd say the difference is reasonably not statistically significant.

    [​IMG]

    The most important fact is that all the rotations start and stop simultaneously, and there are three separate phases of rotation in this slice of the footage. That indicates a common cause.
     
    • Like Like x 2
    • Informative Informative x 1
  13. Gerard

    Gerard Member

    I've been away from this discussion for a while and may not be fully up to date on everything but igoddard's work with the rotations certainly looks like strong evidence that the rotation effect is coming from the camera side.

    One thing I'm wondering about are the differences between the white-hot and the black-hot sections of the video. The shape of the object looks significantly different between these two sections and the pixel values are different as well.

    Examples:

    frame 100:
    obj100.

    frame 700:
    obj700.


    In the black-hot frames most object pixels are fully saturated (ie. 0) while in the white-hot frames they are not. An example value is 162 (out of 255 max).

    I wouldn't normally expect switching between white-hot and black-hot modes to cause changes like these. Any thoughts on what might be going on ?
     
  14. Mick West

    Mick West Administrator Staff Member

    There's two things. Firstly those frames are really far apart. If you want to compare the modes, then do it when the camera actually switches.
    Metabunk 2019-08-07 07-55-13.

    Secondly, the white hot is NOT just an inverted black hot image. If we invert frame 372, we get:
    Metabunk 2019-08-07 07-57-26.

    So it seems clear that brightness and contrast adjustments, are done after the raw image is inverted.

    And if you try to force the image more towards the same brightness curve, you'll see the shape has not really changed at all.
    Metabunk 2019-08-07 07-59-17.
     
  15. Gerard

    Gerard Member

    Can you please describe exactly what operations you performed to get that last result ?
     
  16. Mick West

    Mick West Administrator Staff Member

    Invert, then
    Metabunk 2019-08-07 09-38-23.
     
  17. igoddard

    igoddard Active Member

    Thanks Gerard! I'm just exploring further into Mick's observation of 'bands' across the background that rotate in coordination with the thermal signal of the Gimbal target...

     
  18. Gerard

    Gerard Member

    I don't suppose you'd know how those parameters are defined in the software you're using in terms of their effect at the pixel level ?

    I'm writing some analysis code for this video and that's one effect I'd like to replicate.
     
  19. Mick West

    Mick West Administrator Staff Member

    No, it's just Photoshop CC, and the standard Brightness/Contrast adjustment layer.