Camera tracking: Jitter


I made this for getting deeper into camera tracking with Cycles. If you look closely, you see some jitter of the Quadbot in the beginning. However, when I play the animation inside Blender, there is no such jitter. It seems to be a rendering artefact. Is there any way to prevent those? I tried activation AV-sync, but this reigns havoc on the border of the shadow catcher in the alpha channel…


I can’t see any jitter, looks like a clean track to me

It’s in the first 4 or 5 secs. It’s not big in the half HD rendering but in the 1080p rendering it was so obvious that I discarded that. I got a solve error of 0.77-something. From what I read, that’s pretty ok.

Ok, looking at it a little closer, I can see a bit of shifting in the first few seconds. What kind of camera did you use? Is it possible you’re getting some rolling shutter artifacts?

Canon EOS 6D with 24-105L f/4 IS. I wasn’t moving that fast, I think. The strange thing is that I don’t think I see this shifting when I play the anim inside blender in camera view.

Is your framerate matching in your 3d scene and in your original footage?

Jup, both 25fps.

Maybe I should just file it under “strange stuff happens” and do some more tests with a new video and new scene…

You should go to the movie clip editor and enable the graph view to see where is the jump happening, that will tell you which tracking point is generating a wrong solver at that point, then you have to options: you can re-adjust that point manually to correct the position or you could delete that frame and allow some interpolation between frames, that helps sometimes to have a more smooth solution even if it’s not 100% accurate. Finally I suggest to see the values you get from your tracking marks to see which one is causing the problem, sometimes adding other marks or removing the ones with more errors helps to get a better final tracking calculation. Remember that is important to put markers in all the possible distances in order to reconstruct the scene in a much better way.