-
Are the distortion categories used by RC - Brown 3 or 4, with tangential2, K+, Division - defined/documented anywhere? Googling for these terms produces nothing. What would Brown1 and 2 be? i.e. is this a quality scale? Likewise, what would tangential1 be?
-
If documented, how do we determine which is correct for a particular camera?
-
Doesn’t RC detect the distortion model automatically for each photo?
Hello Tom,
you can find more information in application Help - Learn more: - Alignment settings - Lens distortion models. You can choose Prior lens distortion - model in Selected input table after selecting image/images.
Thank you - I did read that, and the Wiki article linked to, and more.
Are we saying that we just have to try the different Alignment models that RC offers, see which looks best? Aided by various guidelines.
But that there’s no way to look up the correct distortion model from published data for the particular camera and/or lens?
What about Q3 - Doesn’t RC detect the distortion model automatically for each photo?
Hello Tom,
You can set the prior distortion model for each image in Selected input table in ‘Prior lens distortion‘:
“Prior” – here you can define let’s say the weight of the prior information about distortion model.
If you set this value to ‘Unknown’, RC will define the best fitting distortion model for each image. If you expect large distortion, it is best to set this value to ‘Unknown’.
If you set this value to ‘Approximate’, you can define the approximate value of the distortion parameters in the model in which you know these parameters. In this case RC will use these parameters but these can and will be adjusted to best fitting distortion.
If you set this value to ‘Fixed’, then RC will use exactly those parameters you set or will not calculate any distortion if you set ‘No lens distortion’. Use this option only when you are certain about the prior information of the distortion model.
By default, the lens distortion model is set to “No lens distortion” with prior set to “Approximate”. This means that RC tries to find a solution where lens distortion is as close to zero as possible. This is not realistic if images suffer with a visible lens aberration. In such case, you should either define proper distortion coefficients or set the lens prior to “Unknown”.
“Model” – all the listed distortion models are commonly used. No lens distortion means that no model is applied.
Division model is one parameter model used for simple distortions.
Brown3 is the most popular polynomial division model. Number 3 means that three parameters are used for modelling radial distortion. Brown4 means that four parameters are used for radial distortion. Brown4 can be considered as more precise and is used for more complex distortions. In both Brown3 and Brown4 no tangential distortion is modelled.
Brown3 and Brown4 with tangential means that in addition to radial distortion, tangential distortion is modelled as well. RC uses only 2 parameters for modelling tangential distortion.
You can also set the distortion model in Alignment Settings - Distortion model.
Here you can define the distortion model of a created component. You can see the parameters of the distortion model when you select the specific image in the Selected input table – Registration lens distortion. In addition to previously mentioned models, here you can also choose K + Brown3 with tangential2 or K + Brown4 with tangential2. These models add to the Brown model a possibility to optimize the whole camera calibration, including Skew and Aspect ratio.
I begin to understand Zuzana’s explanation, which is so excellently full of practical guidance, it ought to be in Help. So can I ask some more:
“You can set the prior distortion model for each image in Selected input table in ‘Prior lens distortion‘” -
“If you expect large distortion, it is best to set this value to ‘Unknown’” - when would I expect large distortion on a particular photo, if it was taken (along with the rest) with same lens at unchanging FL and known distortion - in my case
If I should be confident of same distortion for all photos, where would I get those figures to “define the … value of the distortion parameters in the model in which you know these parameters”? The parameters offered are FL, Principal point x and y, Skew, Aspect ratio, and the 6 options under Model and Radial 1 2 and 3 - but DXOmark gives as data only 3.12% long and 1.3% short edge (surely lens data doesn’t get better than DXOmark?). And I would have to enter figures for every photo one by one?
What is the advantage of entering these figures and setting ‘Approximate’ or ‘Fixed’, rather than ‘Unknown’ and letting RC “define the best fitting distortion model for each image”? Is it a similar advantage to setting Import settings>Group calibration by exif?
Lots more to ask!
So I run Alignment and now there are new figures.
In 2Ds, a typical registered photo shows pp: -11px, -23px; lamda: -0.092, 0.017, 0.006.
And in 1Ds Reg calibration>Principal point x -0.01, y -0.215; Reg lens distortion Radial 1 -0.092, 2 0.017, 3 0.006.
So now I know what pp and lamda mean, but pp is in different units (ratio 106) compared to Principal point.
All these figures are identical for all of the photos registered in the Component, but have changed relative to Prior calibration>Principal point x 0.00 and y 0.00. What exactly has changed and why identical?
I’m quite sure that I’ve seen similar figures, after Alignment, presented as x = 1, y = 1, z=0.8 but can’t find that now. This would seem to relate to the DXOmark lens-test figures of long-edge (x?) 1.3%, short-edge (y?) 3.12%. But what can the z be?
Hello Tom,
you do not need to enter the calibration and distortion parameters manually in the application, you can write them in XMPs and load them along with the photos. You can find more information for example here:
https://support.capturingreality.com/hc/en-us/community/posts/115001359812-Importing-Camera-Calibrations?page=1#community_comment_115000616692
If you wish to import verified camera calibrations and lens distrortion coefficients calculated outside RC, then use the ‘Fixed’ option. Group images by calibration or lens if you know that a set of images have the same parameters, Rc will calculate the same arameters for the specific group of images.
In the 2Ds view, the principal point x and y are shown in pixels. In the Selected input table in the 1Ds view the principal point x and y are show in mm w.r.t. 35 mm film format. If you export the XMPs for the aligned images, you can see the relative offsets from the center of an image as PrincipalPointU and PrincipalPointV. The values in XMP are the values in the application devidied by 35 mm, therefore dimensionless. Do your images have the same calibration group (other than -1)? If yes, then these values should be the same for each image.
I am not sure what values “x = 1, y = 1, z=0.8” do you mean. You can find these exact values in the Alignment settings (if you use default settings) in the Camera priors table. However, those values specify the precision of the GPS coordinates in each axis.