Reconstruction from unrectified images

[Elmir Jagudin]

Hi

Does colmap support building a reconstruction from a set of unrectified images?

I created a reconstuction from a set of unrectified images from a drone, but it comes out curved.

Looks like this:

This is from a level flight over the area, so the camera positions should all be in the same plane.

Is the curvature caused by the fact that colmap does not take into images need to be rectified? Or is this
coused by something else?

The reconstruction is build with follwing commands:

 colmap feature_extractor --ImageReader.single_camera yes <..>
 colmap spatial_matcher <..>
 colmap mapper --Mapper.ba_refine_focal_length no

Best Regards,
Elmir

[Graham]

I think you are probably running into the same issue mentioned here: https://groups.google.com/d/msgid/colmap/6654721e-d5be-4fa9-b758-5a3f33e007f7%40googlegroups.com

[Elmir Jagudin]

Not sure that it's an issue with automatic calibration of intrinsics.

As far as I can tell colmap is using the camera instrinsics from the JPEG's meta-data. Also, in cases when I photograph the similar scene in 'rectified' mode on the same drone, that is the drone rectifies images on-board, the reconstruction come out correct.

Johannes, can you clarify if the colmap supports reconstruction from unrectified images automagically? Or is possible to specify dewarping correction parameters somehow?

Thanks in advance,

Elmir

[Pierre-Olivier Agliati]

Hi,

Another possibility is that the distorsions of your images are too strong to fit the camera type SIMPLE_RADIAL.

I typically encountered this kind of problem with the GoPro mode called LINEAR which I supposed to be linear.

I solved it by using a camera type with more parameters (I believe it was RADIAL_FISHEYE), what removed

the curvature.

I take the opportunity of this thread to notice that a reconstruction that could be generated with a simple camera type

(like the SIMPLE_RADIAL) may fail with a more powerful camera type extending the first one. This looks a little 

weird to me because the first solution could be theoretically found with additional parameters set to 0, and I was thinking

that the additional parameters could somewhat help to improve it, but I guess that Ceres can't manage to converge 

to something close to this first solution, maybe because of the initialisation of these parameters ? 

Any clue is welcome...

Best regards,

Pierre-Olivier

--
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/9c0d15df-8f17-4b0c-b139-b2922a155f9a%40googlegroups.com.

[jerb...@gmail.com]

Just to jump in here and say that setting the lens to RADIAL_FISHEYE fixed the curved/bent results from my recent planar scan.

Here's an image comparison of the scan of a deck of a ship.

Thank you for the tip!