The intrinsic parameters matrix

[Gwo Jong Huang]

Hi!

My instrument use 5 cameras to capture the snowflake from different angle.  We have fixed focal lens (12.5 mm for all cameras) and square pixel ( 2 cameras have 3.75 um  CCD and 3 have 3.45 um).  I set "config.cal.SQUARE_PIX = 1" in "configdata.m",  and the self-calibration code did not give good results.  So I change this setting to 0 and get good results (at least in the output figures).  I decompose the P into K (internal parameters matrix) and R (the rotation matrix) and then compute T (translation matrix).  Now I have two focal length (K(1,1) and K(2,2) ).  The ratio of two focal length (alpha) is from around .92 to .86 (each camera has different alpha).  The K(1,1) I got and convert to mm are from 12.4 mm to 14.5 mm.
I also get the principle points outside of picture (negative value).  My 3-D reconstruct software assumed square pixel (So only one focal length).  I use the mean of two focal length and the principle points I got (even some of them are negative) and run 3D-reconstruct code.  I got the shape about right but the size are wrong.  I found that the size can be change when I proportionally change the translation matrix.  My questions are:
1.  Can I use a known K to run the self-calibration code?  If yes, how to do it.
2, What is the unit of translation matrix?  Is it pixel, m or mm?
3. Is negative principle points reasonable? 

Thank you for help.

Huang    

[Tomas Svoboda]

>
> On 12 Mar 2016, at 02:45, Gwo Jong Huang <gwojon...@gmail.com> wrote:
>
> Hi!
>
> My instrument use 5 cameras to capture the snowflake from different angle. We have fixed focal lens (12.5 mm for all cameras) and square pixel ( 2 cameras have 3.75 um CCD and 3 have 3.45 um). I set "config.cal.SQUARE_PIX = 1" in "configdata.m", and the self-calibration code did not give good results. So I change this setting to 0 and get good results (at least in the output figures). I decompose the P into K (internal parameters matrix) and R (the rotation matrix) and then compute T (translation matrix). Now I have two focal length (K(1,1) and K(2,2) ). The ratio of two focal length (alpha) is from around .92 to .86 (each camera has different alpha). The K(1,1) I got and convert to mm are from 12.4 mm to 14.5 mm.
> I also get the principle points outside of picture (negative value).
> My 3-D reconstruct software assumed square pixel (So only one focal length). I use the mean of two focal length and the principle points I got (even some of them are negative) and run 3D-reconstruct code. I got the shape about right but the size are wrong.

the overall scale is always undetermined unless you provide some additional information, see the info about alignemnt with world (in papers, report, ./LocalAlignment directory)

> I found that the size can be change when I proportionally change the translation matrix. My questions are:
> 1. Can I use a known K to run the self-calibration code? If yes, how to do it.
> 2, What is the unit of translation matrix? Is it pixel, m or mm?

nothing from that. It is simple a unit.

> 3. Is negative principle points reasonable?

the plus/minus in the equations can in some cases cancel out. not sure at the moment. More important is the location. For usual cameras it should be close to the image center.

Tomas

>
> Thank you for help.
>
> Huang
>

> --
> You received this message because you are subscribed to the Google Groups "MultiCamSelfCal"

----------------------------------------------------------------------
Tomas Svoboda mailto: svob...@fel.cvut.cz
Center for Machine Perception http://cmp.felk.cvut.cz/~svoboda
Department of Cybernetics http://cyber.felk.cvut.cz/people/svobodat
Faculty of Electrical Engineering phone: (+420) 224.35.74.48
Czech Technical University in Prague fax: (+420) 224.35.73.85