I’m on my fourth shot of the day to track and I’ve had the same problem everytime. I successfully track and solve the clip with a solve error of about 1. Through the camera, all the points line up nicely. However, markers that should be farther away are the ones that come out closer to the camera, and vice-versa. Everything comes out backwards. Also, the camera ends up upside-down to the grid floor. I have also tried the set scale option to get this to work.
Hi Daccy, the problem outlined in that other thread - not enough paralax may certainly be the issue with a couple of the shots. However, one in particular, a sweeping aerial shot moving sideways is the real trouble maker. Unfortunately I won’t be able to re-shoot it either. I’m stumped that with all four shots I am getting the exact same problem of Blender inversing the close points and the far points.
A solve error of ‘1’ - which is rather high - might have something to do with it. It could be a result due to bad trackers or incorrect camera data, but it’s hard to say without further info. Could you post the blend?
I would agree with Daccy, we need more info and I am pretty sure you have bad camera data. You can get a low error(1 isn’t all that bad) with a completely wrong focal length/sensor size ratio and get a messed up solve like this… I have run into this trying to track pd video with no camera data available including some aerial shots
Well sure, it isn’t terrible. I mean I did a shot a while ago where I managed to bring the solve error down to ~0.9, and it was very stable. On the other hand I did another shot where ~0.4 didn’t work at all, so I had to lower it further. So it kind of depends on the shot I suppose. As you said, much of it is caused by bad camera data.
You say aerials, did the lens have stabilizing? I tried an aerial that was stabilized and could not get a solve due to distortion and changes to sensor centre in the source footage.
I got the footage from bottledvideo.com and according to the site, they were shot with a gyro stabilized CineFlex V.14 so they are camera and not lens stabilized. I only tried a few and the ones that I tried and failed to get a good solve on I figured they had to be partly zoomed and not just camera motion. The sixth one down on this page tracked beautifully
(got tired of seeing tests with holes in the ground so I put one in a hanger ;))