How to provide constraints on image poses?

[SK12345]

Hello,

I was wondering if there was a way to guide the image matching due to some known constraints for the camera orientation?
I'm using a DJI Phantom 4 with a gimbal. Unless I'm in the process of crashing, it is a solid assumption that the camera orientation has near zero roll and pitch is somewhere between looking straight ahead or down. I am getting misaligned images which could potentially be avoided if I could set those constraints somehow.

Any suggestions would be greatly appreciated!

Thanks again 

EDIT: I should also mention that the roll, pitch and yaw values are provided along with the GPS information. Perhaps there is a setting I've missed which allows us to set how much we trust this information?

[SK12345]

Bump. Despite the reference error for GPS and orientation being set quite low, Metashape seems to ignore these settings and severely misalign photos (upside down or sideways in some cases).

[Alexey Pasumansky]

Hello SK12345,

You can enable "load camera orientation angles from XMP meta data" in the Advanced preferences tab before adding the image to the chunk.

Then you can also specify the higher accuracy for the camera orientation angles (from the Reference pane settings dialog or using Set Accuracy command from the Reference pane context menu for the selected cameras). But make sure that there is no incorrect information loaded, otherwise, with the high accuracy the invalid values may affect the alignment results.

The problem with the alignment that you are facing could be a result of the lack of the image overlap, can you share a few screenshots demonstrating the issues with the alignment?

[SK12345]

Thanks again Alexey.

When they load correctly with orientation angles, should the camera positions in the model still be just points (without indicating orientation like it does after alignment)? In the reference pane, the orientations appear to load in correctly.

I've also recaptured my dataset with slightly better images and a different coded target approach and had more success.


P.s. Not related to this issue, but I just wanted to say fantastic work to whoever fixed the running out of RAM problem in one of the last few versions of Metashape. For a long time I couldn't run higher resolution reconstructions even with 32GB of RAM. Now it's been happily crunching away big datasets without crashing. Previously I had many occasions when I had been running a reconstruction for a long time only to have it crash right near at the end. That fix has got to be one of the biggest improvements of late (assuming the fix was with metashape and not a driver issue).