Forum MicMac
http://forum-micmac.forumprod.com?sid=18ddcc204f781d0d947d5b43b0c53a3f

Camera calibration
http://forum-micmac.forumprod.com/viewtopic.php?f=30&t=1957
Page 1 of 1

Author:  20cent [ 12 Sep 2020, 09:52 ]
Post subject:  Camera calibration

Dear MicMac users,

trying to fine tune my camera calibration (Mavic Pro 2 Hasselblad camera), I am expecting to find coherent results between two approaches :
1. gave a try a the Camera calibration toolbox for MatLab ;
2. performed several Micmac runs on a set of images.
The main issue I cope with is, I guess, a units concern : while focal length, PP and Cdist are similar, I am having trouble with distortion parameters, i.e. they differ from each other by several (well, many) orders of magnitude, e.g.
AutoCal_Foc-10790_Cam-Hasselblad.xml :
Code:
...
<CoeffDist>-8.68814596800474489e-09</CoeffDist>
<CoeffDist>1.03782603772602363e-16</CoeffDist>
...

Calib_Results.mat :
Code:
...
%-- Distortion coefficients:
kc = [ -0.131544042569461 ; -0.044695677971104 ; 0.001131062101727 ; -0.000547469947769 ; 0.000000000000000 ];
...


Any advice ?

Vincent.

Author:  Luc.Girod [ 16 Sep 2020, 11:13 ]
Post subject:  Re: Camera calibration

Hei,

Look up this : https://www.mathworks.com/help/vision/r ... nsics.html

And the MicMac doc section 15.2.2

Seems to be the same formulas (as expected). But it seems that the Matlab option gives you the inverse distortion ( Distort=f(ideal) ), while MicMac gives you the direct ( ideal=f(Distort) )

Hope this helps!

Author:  20cent [ 16 Sep 2020, 12:42 ]
Post subject:  Re: Camera calibration

Thank you Luc for your reply, I'll look into this.

I recently stumbled on a notation correspondence between the Matlab calibration tool I use and some previous work by Heikkilä et Silvén :
http://www.vision.caltech.edu/bouguetj/ ... eters.html

looks like I should divide kc(1) and kc(2) respectively by f^3 and f^5 to get back k1 and k2 coefficients: in my set of data, values then still differ but orders of magnitude fit quite well. Did not go further for now.

On another hand (sorry for my ignoramus questions), do you consider it's fondamentally a good practice to calibrate a camera, or do you consider it's ok to run micmac on raw images? The point is Mavic Pro 2 perfoms its own distortion correction while translating raw (.dng) images to jpg format, which I guess is certainly confusing for micmac calibration routines, isn't it?
Is it worth trying to (i) calbrate my camera (ii) un-distort raw images out of micmac (iii) run Tapas process on the latter, relying on a set of parameters corresponding to an ideal camera?

Vincent.

Author:  Luc.Girod [ 24 Sep 2020, 14:24 ]
Post subject:  Re: Camera calibration

Hei,

Tapas should be fed the rawest image possible, as any previous correction might actually create some weird edge effects that are harder to fit than the raw distortion pattern. Pre-correction should only be considered if the distortion pattern is known to be very non standard (like if the camera is not using a standard lens but something more exotic). Tapas includes a very large variety of options to fit even some rather weird distortion patterns (fish eye, non centric...).

Author:  20cent [ 25 Sep 2020, 15:05 ]
Post subject:  Re: Camera calibration

Thank you Luc for these clarifications,

V.

Page 1 of 1 All times are UTC + 1 hour
© phpBB Group