Question about the PCA for spatial-variant data

[ewrfcas]

Thanks for your good work!
Here is a question how to use PCA to achieve spatial-variant data?
Because the colmap coordinates are somehow random, while the camera would not locate in a certain plane (xy or yz) as shown below:

Maybe all three coordinates (xyz) enjoy large eigenvalues, while only one plane should be considered (tilted plane).
So why could using PCA with XYZ get the correct projection distance D? How do you set the dimension in PCA (2 or 3)?
Thanks.

[heheyas]

Hi, ewrfcas, checkout here for a more detailed camera pose orientation and centering.

[ewrfcas]

Thanks for your reply, I would try this function.