Forum

Author Topic: HELP!> Good photo coverage and bad alignment  (Read 2463 times)

HCALDERON

  • Newbie
  • *
  • Posts: 1
    • View Profile
HELP!> Good photo coverage and bad alignment
« on: August 01, 2018, 10:06:33 PM »
Dear Friends,
 
We have a project with near 1750 photos. Our area study is a 30 Km inland water canal and our final product is creating a 3D model of all the walls of the canal in a seamless format.
All of our photos are geotagging properly and geolocated in the right spot which allows us to ensure that we have enough coverage and overlap between the 2 neighbors photos.
After running the Agisoft Photoscan we realized that some of them are not properly aligned by the tool. However we check the coverage and they have an average of 60% overlap.
There were several (most) photos that did not align even though it was clear that there was lots of overlap AND they were positioned correctly.
So, we took different chunks with about 5 of them in each to see if Photoshop could mosaic them automatically, and it did. However it did not align in Photoscan.
Sample Data:
121 geotagged photos (https://we.tl/9AuyNEcy0S)
 
Procedure:
We followed all the step as per the user manual is different settings and still is not working.
It could possible that we are missing one the steps but so far we confirm that all the steps were done properly and still without any success on this process.
The final ideal scenario is to be able to use the 3D model in .PDF format to attach it in digital HTML report format, like we did in some other areas

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14847
    • View Profile
Re: HELP!> Good photo coverage and bad alignment
« Reply #1 on: August 08, 2018, 03:01:37 PM »
Hello HCALDERON,


It seems that the image matching operation produce only a few tie points due to the complexity of the surface being scanned:
the branches of the trees and bushes produce different pattern even for the neighboring camera locations, so PhotoScan fails to match such areas. I assume that the result could be better, if the hill slide is visible on the the images - then the matching points related to the bare ground would improve the stability of the alignment.

So the problem is not related to geotagges and the main reason of the low number of detected tie points is complexity of the surface structure, even with such overlap ratio.

Probably, taking images in the landscape orientation or with even wider lens will help to get some matching be the sides of the images, where the difference in the branch pattern is not so considerable.
Best regards,
Alexey Pasumansky,
Agisoft LLC