Forum

Author Topic: Processing spherical images  (Read 4343 times)

Paulo

  • Hero Member
  • *****
  • Posts: 1557
    • View Profile
Processing spherical images
« on: February 06, 2020, 07:54:23 AM »
Hello,

I have a set of 116 spherical 32 MPix images stitched from six 5 Mpix images from a camera rig installed on roof of car...

I masked the sky in each panorama with Mask by color script and also masked the bottom of each panorama (car).

I use following alignment parameters

Code: [Select]
accuracy = High, preselection = generic, reference, keypoint limit = 10000, tiepoint limit = 4000, apply masks = 1
But the alignment fails miserably both with Apply masks to keypoints and Apply masks to TiePoints options....

Here is the link to data set https://drive.google.com/drive/folders/1ymkGRrYssREVcLpml_3C1bLTYIKIK9Ud?usp=sharing...

Any help on this will be greatly appreciated,
« Last Edit: February 06, 2020, 08:15:46 AM by Paulo »
Best Regards,
Paul Pelletier,
Surveyor

James

  • Hero Member
  • *****
  • Posts: 771
    • View Profile
Re: Processing spherical images
« Reply #1 on: February 06, 2020, 11:40:12 AM »
Just looking at the thumbnails, it looks like there is nowhere near to enough overlap between adjacent images.

Things that are just coming into view in one image are right alongside you in the next image, and then far behind in the one after that.

The difference in scale and perspective angle mean metashape has no chance of finding a good match most of the time, and the trees in the central reservation mean that there will be next to no matching between the parallel survey lines.

I would suggest increasing the number of images by x3!

Paulo

  • Hero Member
  • *****
  • Posts: 1557
    • View Profile
Re: Processing spherical images
« Reply #2 on: February 06, 2020, 05:42:59 PM »
Thanks, James for the prompt reply!

I guess you are right. The 6  images set are taken every 10 meters and previously I have processed a successful spherical project with images every 3 m!

Best Regards,
Paul Pelletier,
Surveyor