Motion tracking can't get lower error

Hello,

This is a green screen footage on which one I have increased contrast and saturation for better tracking, when I check my trackers, they are perfectly tracked, but when I solve I always get 1.83 error av and if I try to clean up trackers, the error will be .90, but if I keep cleaning it will vanish most of my trackers so I can never get a low average error.
I have tried every A and B keyframe, tried to place trackers on the edges, corners, middle of the green crosses and nothing works then when I check track panel they literally don’t move.
Motion model is loc and scale, correlation .9, 5 pixels bordel, what am I doing wrong ?

Dont know, best to post the video for others to try.

1 Like

I have uploaded the image sequence in my google drive, if you wanna try, I can’t get lower than .90.
If you try i’d like you record your screen to show me please

https://drive.google.com/drive/folders/1SgBJ1WfixHn3QbSzvpG_2i_9pIaefwxk?usp=drive_link

Will try, maybe in the mid day break or after work.

1 Like

Thank you.

Try Refine Tracking Solution Clears errors.

Refine Tracking Solution — Blender Manual

If the solution is somewhat right then the addon can work… If its totally bad, not the right perspective and so it wont work.

Using an addon is okay but this shot is very simple to track so I am really wondering what’s wrong with the solving, my trackers do not move at all, even in graph, but whatever I do, the error will always be 1.28 and if clean up I can reach minimum .90 before it vanishes everything

Check this out, What do you think ?

1 Like

That looks pretty perfectly. Its a long panning shoot… The tracker is sometimes hit and miss with the solves but yours work.

Yeah, this is what i’ve come to realise, because graphically it displays a high error but… It works lol

Set Keyframe B to be the area with the most motion. Where the red lines are highest or lowest. The A & B location / rotation need to be as different to each other as possible.
You can also tick the Keyframe Radio Button just above the A & B keyframes and let Blender pick the frames with the best affine difference.
Click the Tripod Radio Button if the real life camera is on a tripod.

You problem is probably a lack of depth information. I am going out, so I dont have the time now to look at your footage. It always helps the solution if you move to one side at least 45 degrees as part of the footage to create triangulation. The Red and Green lines show the camera has not moved much. Looks like a pan with the person.

1 Like

I have tried but I still get the same result, I can see that the solving isn’t perfect, I really need someone to show me his interface and get the low error solving :frowning:

it could also be rolling shutter error

There are several motion-tracking lectures on the link.
It seems impossible to find lectures that implement the same environment.

https://www.youtube.com/playlist?list=PLp90cx0wSyKa4OELMsrhfTymHggizAIak

… If it is the footage that is the issue, then lessons about the interface will make little difference.

Have downloaded the footage. I was right - it is a constant pan. Blender is not good at solving this kind of footage. Wrong algorithm.
I am not going to even try tracking that. Sorry. And too make it worse - its Jpegs, so every new save is downgrading the images.

Have you tried doing a couple of Plane Tracks? One on the floor & one on the wall. That may just work.

The error in the attachment is 0.16.
But I don’t think it’s a very good result.

I don’t think the error rate should be low to show good results. :sweat_smile:

dfs.blend (1.3 MB)

Maybe, when I have exported my footage from after effect, I have added a warp stabilizer and there is maybe a distortion up and down, i’ll try without

The error is very low but it seems to be shaking

Yes, the error rate is low, but you may think that the wrong value is perceived as good