I'm in Movie clip editor, blender v2.77 running on xubuntu 16.04.
I've taken video footage of a cylinder (a 1.5l drink bottle wrapped in paper) with ample track points marked all around the cylinder. In the video (from stationary camera) I manually rotated the cylinder for a couple of rotations. I tracked plenty of points (20+) as they came into view on the cylinder's surface, and disabled tracks as they rotated out of view. The same track was reused for the same point as it came into view a second/third time as the cylinder was rotated.
when trying to solve for the object, I'm getting continuously very high errors of 57-ish with no [major] errors in the tracking points. The 'camera' was set up correctly (gopro 3 black) and I adjusted the k1 value (using the grease pencil/'calibration' feature of the 'display' section) and even added markers for the 'camera' and tracked the (mostly) very stationary background and solved the camera movement (but obviously didn't get any 3d data out of the camera solve), error < 0.4.
initially, I had a few errors in the tracks with some tracks still 'tracking' when the tracking point went behind the cylinder. I spotted these on the 'graph clip editor' and corrected them, but this didn't improve the object solve error at all/very much.
So, my question is; What else could I do to try and track this rotating object accurately? Can blender even handle this kind of tracking scenario? Should I instead treat the cylinder as several different faces and track each face as a separate object? other ideas?
I've just a ran quick test where I followed my own advice of tracking a single 'face' of the cylinder over a small rotation. That solved to a error of 7 (still high but more expected) so it seems that blender may not be capable of tracking a fully rotating object, but I'd still like appeal for help or confirmation.
(btw ,i'm new here, please be gentle... :P)
EDIT: I've uploaded my project to drop box available til end-of-november 2016 (link:https://www.dropbox.com/s/kx6bwzeuxmpimuk/spintest.zip?dl=0). looking at the 3d view of solved object, the tracks judder everytime a marker becomes 'active' again after being disabled whilst the marker is behind the cylinder and there is no obvious 'rotation' of the solved object. I expected to see all the tracks of the object in a cylindrical shape.
The purpose of this project was to test blender's capabilities when it came to object tracking, but if blender is lacking in this type of tracking (rotating object from tripod), then how should I go about tracking a rotating object?