Forum

Author Topic: 360 Panorama alignment to world  (Read 2164 times)

spader

  • Newbie
  • *
  • Posts: 8
    • View Profile
360 Panorama alignment to world
« on: November 11, 2021, 05:10:42 PM »
The panoramic results from MS are great, but lacks automatic horizon alignment. Aligning the horizon has to be done manually, and as far as I can see, can only be done on on single-image chunks. The same holds for aligning images to north.

If, on the other hand, we use multiple camera stations that are aligned with eac hother and perhaps a few GCPS, then it it is not possible to rotate object as is suggested in the tutorial at https://agisoft.freshdesk.com/support/solutions/articles/31000148830. It is currently possible to align multiple camera stations with each other, and thus also with GCPs. Therefore, the exterior orientation of the panorama should be known. Yet, when I try to output the results, the panorama is randomly rotated.

I thus suggest the following feature. If the ingoing spherical panorama has a known exterior orientation, then allow us to export the panorama perfectly levelled and with the equirectangular center pointing exactly north. This would allow usage of Metashape for perfectly georeferenced 360 dataset generation.




spader

  • Newbie
  • *
  • Posts: 8
    • View Profile
Re: 360 Panorama alignment to world
« Reply #1 on: December 05, 2021, 11:44:53 AM »
The  1.8.0 build 13597 was released since my last post, with the excellent addition of "Added Build Panorama command with seamline editing support". This is great news! As panoramas are apparently being worked on at this moment, I would like to add a few more thoughts on how its usability could be improved.

Suggestion #2: Allow for precise angular alignment of a 360 using markers in an image with n projections = 1. Lets say you have a group of images marked as a station and nothing else in a chunk. The images are aligned, and the tie points appear as the classical spere. My suggestion is this: allow adding markers with known reference in a single image. Then the direction of that ray is known. Then add a few more markers with known coordinates, yielding more rays. Optimizing with multiple rays should lead to that the orientation of the spherical panorama should be pretty precise, as well as yielding some scaffolding to avoid drift.

Use case:  we have a drone pano shot above a city. Mark three or more house corners, and get their reference XY positions from a typical map service.  Optimize. Build Panorama. The direction that this resulting panorama looks in should now be in relation to the world, for example that the middle of the equirectangular image point straight north.

Suggestion #3: Assume that we have multiple previously created 360 images using the Build Panorama. These are loaded as single images into a chunk, and their projection set to spherical. Allow for a common alignment without calculatin tie points by simply marking common points. Then allow export with exif tags as suggested in #4.

Suggestion #4: Export the correct exif tags of the spherical image. Approximate the center of the spherical image from ingoing image coordinates and use it as exif coordinates. Do the same for yaw/pitch/roll.

Why is this, and my last suggested feature, useful? 360 images are increasingly used for VR/AR/BIM use cases. Being able to precisely align these panoramas to world, in an overfit manner, either by themselves or in combination with others, is critical for this use case. We can get a good enough approximation of the camera position using PPK or RTK, but the use case critically need a bit more robustness and precision in relating the image alignment to the world system.