Hello,
In my journey to learn photogrammetry with RC… What is the best camera distortion model to choose ? I guess that this depends, but what would be the good workflow to choose one or the other ? I already read the documentation that explain well what these models are, but not really how to choose them.
Should we ran tests with the same dataset (pictures) and alignment settings and look at the error or the total projections count with each distortion model ? and take the lowest one ?
My idea is that this choice depends on the lens so should be done per lens and camera ? is this right ?
Once I have a camera distortion model, should I keep it all the time (with this camera/lens couple) ?
Or is there a camera distortion model that is really better than all the other (but maybe more processing intensive ?)
Thanks for any useful information.
Hi Jonathan
there is a simple rule, but not always 100% accurate.
Brown3 or Brown4 for most cameras ( APSC-FF-MediumFormat-LargeCam )
Brown with tangential is good for small sensors, like compact cameras, as they are of poor design and quality so any additional parameters need to be calculated to compensate for the wrong results.
Division is good for fisheye lenses.
whats the best way to deal with projects with mixed lenses?
ie fisheye, that you’d want to use division for.
and standard lens that works better with brown3.
Hi Chris
again, it depends on more factors so there is NO definitive rule what to choose… I would align it as separate COMPONENTS and just then get all of them into the same project, but ONLY if I was experiencing issues with alignment.
If you capture a project properly ( enough overlap ), then Brown 3 is sufficient, even if there are fisheye lenses…
And is this better to group the calibration if they are all taken with the same lens ? (“Set Constant calibrations”)
Does focusing have an impact on calibration ?
I am resurrecting this thread.
So, what lens distorsion model do you use and in what conditions ?
Has one proven to work in most cases ? Milos is often using the “brown3 with tangential 2” in his settings. It seems it is a good setting for most cases ?
What can we compare to benchmark the best model to use in a particular case ?
I started to do some comparisons, but results are hard to analyse : sometimes the same settings give 2 different results (more images aligned the 2nd time). I think this is due to the way the RC algorithm work (using previous calculation results).
Hi Jonathan,
as Wishgranter said, there is no single answer. It depends on the camera and the object and the images.
So you need to try it out. I used the last in the menu quite a bit and got good results. Mine is a small, cheap camera, so that confirms what WG said.
Just use a small fraction of your images of one spot (say 50) and make comparisons.
And yes, RC DEFINITELY uses information from previous alignment, IF they are present as components.
But I also am convinced that there is a randomness in the algorythm somewhere, so each alignment will lead to slightly different results. I am talking about internal mechanisms though, not necessarily noticeable differences. Altough even that I have observed.
I think that the technology is not something that has one right method, but it needs experience to get a feeling for when the results are as good as they are going to get.
Hi,
This is an interesting thread. I am modelling a wall with laser scans that are registered perfectly, GCPs extracted from the pointclouds and then photographs taken with a Fujifilm X-T3/27mm lens.
Despite importing the laser scans as exact/georeferenced AND fixing the prior pose as Locked, the images and scans were deviating wildly depending on various settings. By deviation, I mean to the order of tens of metres over the length of the wall.
It was only when I started experimenting with distortion models that things started to make more sense. I found that K + Brown3 with tangential 2 gave the only good result.
As a side note, RC reports that my camera is not in the database, so part of me wonders whether this is part of the issue?
Does anyone else have a comment on the choice of distortion model for hybrid photogrammetry>