Forum

Author Topic: Difference in camera calibration values  (Read 2289 times)

Jobbo90

  • Newbie
  • *
  • Posts: 19
    • View Profile
Difference in camera calibration values
« on: October 29, 2020, 02:37:29 PM »
Hi All,

Following up on some discussions I found here I started diving into two projects I have over the same area but with 9 months difference between acquisition. Both projects were acquired with the same DJI mavic pro 2 and the same flight paths planned in Lithci (nadir and off-nadir pictures at each location). The only difference is the flight height which was 50 meters in 2019 and 75 meters in 2020 (relative windy conditions, so the ensure sufficient overlap I dicided to increase the flying height). The results showed here contain roughly 50% of the pictures processed inAgisoft metashape 1.5.2 (I'll try 1.6.x soon as well!). All alignments are at 'standard' settings:
high accuracy, generic preselection ON, reference preselection ON, key point limit 40.000, tie point limit 4.000, adapative camera model fitting ON.

I started notcing I considerable doming effect in the 2020 data and found in other discussions here that this might be related to the automatic camera calibration performed by Agisoft. When I look at the estimated calibration camera's I get confused.  So for 2019 all seems to be fine with estimated f at 4289.5 (intial at 4256), and cx,cy at 20 and -49 respectively and b1 and b2 at 0.23 and 0.38. Allthough I wonder why the estimated f is different and if so if the difference is acceptable. (hopefully the example images are clear!)
For 2020 is becomes messy: estimated f is at 3366.8, cx and cy at 63.8 and 1556.1 respectively, and b1 and b2 are -66 and 22.  Remember, these are the same drone and flight lines (except the heigt). So to check if something else changed I looked at another fleight campaign I did around the same date in 2020. Here the camera calibration values are way better: estimated f at 4256 (it is the same as the intial...) and cx and cy at 17 and -27 respectively. b1 abd b2 are at -22 and -0.15, so only b1 is still way off.

So my questions is where these difference come from, is it camera settings during flying or something else? And more specifically what can I do to improve the alignment of the 2020 data?
I already read something about enabling rolling shutter before alignment (this doesn't improve the 2020 dataset) : https://www.agisoft.com/forum/index.php?topic=12645.0
https://www.agisoft.com/forum/index.php?topic=12178.0

Also trying to estimate the camera calibration parameters beforehand and 'fixing' them seems to be an option:
https://www.agisoft.com/forum/index.php?topic=11983.0
Just wondering if that is going to help?

Some additional info:
both datasets are part of a larger set that also contains 20 - 25 GCP markers (dGPS quality) that do help for the overal quality of the outputs. Yet especially the DEM suffers from significant doming in the 2020 dataset, but also the orthophoto is distorted to the large noise, even after gradual selecting tie points with relatively strict reprojection error values, reconstration uncertainty and projection accuracy filters. Also the lay-out of the markers is not optimal due to limited accessibility at the boundaries (muddy intertidal area). This is why i'd like to see if there is improvements possible in the allignment and camera callibration.

Curious to see if there is somebody who can direct me towards a potential solution for the 2020 dataset!

Paulo

  • Hero Member
  • *****
  • Posts: 731
    • View Profile
Re: Difference in camera calibration values
« Reply #1 on: October 29, 2020, 10:36:28 PM »
Hi Jobbo,

please look at following post https://www.agisoft.com/forum/index.php?topic=12471.msg55469#msg55469

Note that you should check rolling shutter compensation in Camera calibration window....

Hope this helps,
Best Regards,

Paul Pelletier

Jobbo90

  • Newbie
  • *
  • Posts: 19
    • View Profile
Re: Difference in camera calibration values
« Reply #2 on: October 30, 2020, 04:54:12 PM »
Dear Paolo,

Thank you for your swift reply, I indeed followed these steps before but now I also tried with the newest agisoft version. It does indeed improve the Cx, Cy and f estimates. Yet I'm still left with a doming effect in my model (sparse cloud) as can be seen from the pictures. Do think this can be improved before or during alignment or is that really only possible with the ground control points? I have them for this project but due to restricted distribution through out the scene I'd like to get the sparse point cloud as good as it gets before I update cameras after including markers.

It seems to me that especially b1 and b2 are causing some problems, I also tried the workflow you suggested in the post by fixing b1,b2 together with f, cx and cy before aligntment. And then during the update cameras steps I include all of those camera parameters, not sure of that makes a difference?

I often follow this workflow: http://www.rslab.se/agisoft-photoscan-pro/ where the markers are introduced before filtering on the reprojection error..

Regards
Job


Jobbo90

  • Newbie
  • *
  • Posts: 19
    • View Profile
Re: Difference in camera calibration values
« Reply #3 on: November 04, 2020, 01:28:48 PM »
Hi all,

I haven't been able to resolve the doming effect that is visible in the previous post. If anybody want to give it a go at the data sample I used, that would be much appreciated:

https://surfdrive.surf.nl/files/index.php/s/D73dTKjcbIYFC2T

I'm curious what others are finding and might have some valuable suggestions to improve for future efforts.

Regards
Job