OutPut camera-pose

[松本慶太]

I estimated camera-pose with COLMAP function.

And I got images.txt noted following link.

https://colmap.github.io/format.html

What are the units of these parameters of the translation vector? 

I'm assuming it's in pixel units, is that correct?

Also, it seems that the estimated translation vector is not a unit vector. Can I assume that the scale estimation is also done at the same time?

[Wayne Cochran]

The camera rotation R is given as normalized quaternion (qw qx qy qw) and the translation
t is a 3D point (tx, ty, tz) that is typically in colmap's normalized 3D coordinate system where

all the point coordinates lie in the range -10 < x,y,z < +10.   

Colmap uses Eigen to convert to matrix form:

Eigen::Quaterniond q = Eigen::Quaterniond(qw, qx, qy, qz);

Eigen::Matrix3d R = q.normalized().toRotationMatrix();

Eigen::Vector3d t(tx, ty, tz);

If you want to retrieve the position of the camera C you can solve the following

RC + t = 0

thus

  C = -R^T t

which will be in colmap's normalized units (not pixels). 

[Johannes Schönberger]

SfM reconstructions do not have scale. The scale does not have a unit but it is roughly scaled to the range [-10, 10] for numerical stability reasons.

--

You received this message because you are subscribed to the Google Groups "COLMAP" group.

To unsubscribe from this group and stop receiving emails from it, send an email to colmap+un...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/colmap/9bb74194-ec03-4f69-ba3a-b5079f92c3cbn%40googlegroups.com.

[松本慶太]

Thanks to both of you.

It has been helpful.

2021年5月8日土曜日 18:36:43 UTC+9 Johannes Schönberger:

[Adam Polak]

Sorry to resurrect this thread but I was wonder if it was possible to find the rotation of the camera poses once you find the camera center? (the center being found using C = -R^T t)

If my question is unclear I made a diagram to help explain: https://github.com/colmap/colmap/issues/1376#issuecomment-1013938947