Generating camera poses subject to known constraints (outward-facing)

[Kyle Corbitt]

Hey all,

I have several thousand videos taken by standing in one spot and turning in a circle with the camera pointing outwards. These videos were originally intended for panorama generation.

I'd like to extract frames from these videos and use COLMAP to determine their exact poses as part of a pipeline. When I pull enough frames from the video COLMAP is able to successfully match features and generate poses, but I find that the generated poses are consistently inverted—COLMAP places all the cameras inward-facing, instead of outward-facing (similar behavior to this issue).

Is there any way to convince or constrain COLMAP to try to place the cameras along an outward-facing cylinder? Alternatively, is there a way I could pass my "best guess" initial poses to COLMAP and have it fine-tune from there? That might be good enough to snap it into the correct local minima.

Thanks so much!

Kyle

[Wayne Cochran]

Since the camera remains in the same spot there is no parallax so Colmap will struggle to localize anything in the scene. Read up on best practices for photogrammetry and the first rule is to never take photos from the w same location.

--
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/cc8ef806-c104-472e-97bc-22340c9d1c8dn%40googlegroups.com.

--

Wayne Cochran

Computer Scientist

way...@quintar.ai

[Kyle Corbitt]

I totally understand that I'll get better results with more movement.

The pictures weren't taken from one spot though—users generally hold their arms out, so I get 1-1.5m of translation between captures on opposite sides of the circle. And in practice that does seem to be enough parallax for COLMAP to work with in my case, since it produces an (inward-facing) circle with the frames in the correct order and relative position. I think that if I could just convince it to jump out of the local minimum it's finding and try turning the cameras around it could find a correct set of poses.