Hi,
    Has anyone used Zhang from Microsoft camera calibration software and if so has anyone compared it to the calibration done in AR Toolkit? 
Thanks for your input.
 
Kind regards Danny

Hi,

>    Has anyone used Zhang from Microsoft camera calibration software and
>if so has anyone compared it to the calibration done in AR Toolkit?
>Thanks for your input.

I think ARToolKit calibration method is one of the worst method.
If you can use other methods, it would be better.
It is not needed to compare them.
Because ARToolKit calibration was developed without deep mathematical
consideration.

One reason is that good calibration is not needed for the purpose of
image overlay. But as I know, some people wants to use ARToolKit as a 3D
tracking or measurement sub-system. For this purpose, you need to use good
camera calibration method, good camera and good lens.

I wanted to improve it, but I have not had enough time to do...


-- 
------------------------------------------------------------------
  Hirokazu Kato
  Faculty of Information Sciences
  Hiroshima City University               Phone: +81-82-830-1705
  Email: kato@s ........................     Fax: +81-82-830-1435
  URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/

Hi I am facing problems with the calibration as well. I am using the Vision SDK version of ARtoolKit and I have tried many times to make a good calibration file with no success. I know that there are some commercially available plugins for 3ds max but I have not investigated them yet. Does anyone know another software that could work with the Vision SDK?
 
Thanks,
 
Fotis Liarokapis


----- Original Message ----- 
From: Hirokazu Kato 
To: Danny Nahmias UCL 
Cc: artoolkit@h .................. 
Sent: Thursday, February 27, 2003 11:20 PM
Subject: Re: camera calibration

Hi,

>    Has anyone used Zhang from Microsoft camera calibration software and
>if so has anyone compared it to the calibration done in AR Toolkit?
>Thanks for your input.

I think ARToolKit calibration method is one of the worst method.
If you can use other methods, it would be better.
It is not needed to compare them.
Because ARToolKit calibration was developed without deep mathematical
consideration.

One reason is that good calibration is not needed for the purpose of
image overlay. But as I know, some people wants to use ARToolKit as a 3D
tracking or measurement sub-system. For this purpose, you need to use good
camera calibration method, good camera and good lens.

I wanted to improve it, but I have not had enough time to do...


-- 
------------------------------------------------------------------
  Hirokazu Kato
  Faculty of Information Sciences
  Hiroshima City University               Phone: +81-82-830-1705
  Email: kato@s ........................     Fax: +81-82-830-1435
  URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/

Hi Danny, Hirokazu, and Fotis,
 
I've been through several camera calibration techniques to find that there is no painless method.
 
> Has anyone used Zhang from Microsoft camera calibration software and if so has anyone compared it to the calibration done in AR Toolkit? 
 
Zhang's method is one of the easiest around to use, although I had many problems when using it, due to a severe lack of documentation, especially regarding the orientation of the different coordinate systems used, and the order in which the corners have to be fed to the program.
 
> Hi I am facing problems with the calibration as well. I am using the Vision SDK version of ARtoolKit and I have tried many times to make a good calibration > file with no success. I know that there are some commercially available plugins for 3ds max but I have not investigated them yet. Does anyone know
> another software that could work with the Vision SDK?
 
In order to calibrate a camera, all you need is around 10 to 15 images of a regular pattern, usually a grid, so you can use the Vision SDK or any other software that allows you to grab still images from your camera. You can also use the OpenCV toolbox alone or through Matlab. These last two solutions provide great results.
 
 
 
As for "generating" a proper calibration file for the ARToolKit, I have not done it yet. Basically, it should be a matter of directly dumping an ARParam struct in a binary file (while taking care of endianness). The problem is that I don't know how to match the parameters:
 
dist_factor[2] = 100 000 000 * f
dist_factor[3] = s
 
with the usual 
 
alphau = s * fx
alphav = s * fy
 
and alos with coefficients
 
k1, k3 etc
 
usually used for radial lens distortion correction.
 
Has anyone "converted" the parameters computed from other calibration software into ARtoolKit's ? I know that many people have been "hacking" the default camera parameters, but I'd like to know hoaw to do it cleanly.
 
Hirokazu, could you please explain us how the two sets of parameters relate to each other?
 
Thanks in advance,
 
 
Fivos
 
 
 

-----Message d'origine-----
De : owner-artoolkit@h .................. [mailto:owner-artoolkit@h ..................]De la part de Fotis Liarokapis
Envoyé : vendredi 28 février 2003 11:20
À : artoolkit@h ..................
Objet : Re: camera calibration



Hi I am facing problems with the calibration as well. I am using the Vision SDK version of ARtoolKit and I have tried many times to make a good calibration file with no success. I know that there are some commercially available plugins for 3ds max but I have not investigated them yet. Does anyone know another software that could work with the Vision SDK?
 
Thanks,
 
Fotis Liarokapis


----- Original Message ----- 
From: Hirokazu Kato 
To: Danny Nahmias UCL 
Cc: artoolkit@h .................. 
Sent: Thursday, February 27, 2003 11:20 PM
Subject: Re: camera calibration

Hi,

>    Has anyone used Zhang from Microsoft camera calibration software and
>if so has anyone compared it to the calibration done in AR Toolkit?
>Thanks for your input.

I think ARToolKit calibration method is one of the worst method.
If you can use other methods, it would be better.
It is not needed to compare them.
Because ARToolKit calibration was developed without deep mathematical
consideration.

One reason is that good calibration is not needed for the purpose of
image overlay. But as I know, some people wants to use ARToolKit as a 3D
tracking or measurement sub-system. For this purpose, you need to use good
camera calibration method, good camera and good lens.

I wanted to improve it, but I have not had enough time to do...


-- 
------------------------------------------------------------------
  Hirokazu Kato
  Faculty of Information Sciences
  Hiroshima City University               Phone: +81-82-830-1705
  Email: kato@s ........................     Fax: +81-82-830-1435
  URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/

Fivos,

I don't believe there is a straightforward conversion in the parameters.

Talk to Brendan Hannigan (brendan@c ............), a student in my lab.   
He's been using the "better" calibration algorithms, and is in the  
process of more fully integrating them into the AR Toolkit;  the first  
pass was to use the OpenCV library (I believe) to dewarp the images  
before returning them to ARTK, and then set the ARTK distortion  
parameters to "0".  This has been working fine for us.

He is planning on implementing "observedtoideal" for these distortion  
parameters (apparently it's straightforward) so we can avoid the cost  
of dewarping, but the reverse is not so simple.  He may be interested  
in getting some help (especially on that latter problem, the going from  
dewarped to warped space) ...


On Monday, March 3, 2003, at 11:18  AM, Fivos DOGANIS wrote:

> Dear Hirokazu,
>  
> Thank you for your reply, but I have already read the whole tutorial  
> over and over. I have tries to match the parameters in dist_factor  
> with the "classic" camera calibration parameters used by Matlab and  
> OpenCV. I was assuming that:
>  
> 1-f = k1
>  
> f being the distortion factor in ARToolKit and k1 being the first  
> coefficient of the radial distortion polynomial...
> But my computations only match this equality up to a translation of  
> the centre of distortion.
> Has anyone a clue about this?
>  
> Best regards,
>  
> Fivos
>  
>
> -----Message d'origine-----
> De : Hirokazu Kato [mailto:kato@s ........................]
> Envoyé : lundi 3 mars 2003 16:59
> À : Fivos DOGANIS
> Cc : artoolkit@h ..................
> Objet : RE: camera calibration
>
> Hi,
>
> Hirokazu, could you please explain us how the two sets of parameters  
> relate to each other?
>
>
> Please look at:
> http://www.sys.im.hiroshima-cu.ac.jp/people/kato/etc/ART02/ART02- 
> Tutorial.pdf
>
> --
> ------------------------------------------------------------------
>   Hirokazu Kato
>   Faculty of Information Sciences
>   Hiroshima City University               Phone: +81-82-830-1705
>   Email: kato@s ........................     Fax: +81-82-830-1435
>   URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/
>

Fivos,

I don't believe there is a straightforward conversion in the parameters.

Talk to Brendan Hannigan (brendan@c ............), a student in my lab.   
He's been using the "better" calibration algorithms, and is in the  
process of more fully integrating them into the AR Toolkit;  the first  
pass was to use the OpenCV library (I believe) to dewarp the images  
before returning them to ARTK, and then set the ARTK distortion  
parameters to "0".  This has been working fine for us.

He is planning on implementing "observedtoideal" for these distortion  
parameters (apparently it's straightforward) so we can avoid the cost  
of dewarping, but the reverse is not so simple.  He may be interested  
in getting some help (especially on that latter problem, the going from  
dewarped to warped space) ...


On Monday, March 3, 2003, at 11:18  AM, Fivos DOGANIS wrote:

> Dear Hirokazu,
>  
> Thank you for your reply, but I have already read the whole tutorial  
> over and over. I have tries to match the parameters in dist_factor  
> with the "classic" camera calibration parameters used by Matlab and  
> OpenCV. I was assuming that:
>  
> 1-f = k1
>  
> f being the distortion factor in ARToolKit and k1 being the first  
> coefficient of the radial distortion polynomial...
> But my computations only match this equality up to a translation of  
> the centre of distortion.
> Has anyone a clue about this?
>  
> Best regards,
>  
> Fivos
>  
>
> -----Message d'origine-----
> De : Hirokazu Kato [mailto:kato@s ........................]
> Envoyé : lundi 3 mars 2003 16:59
> À : Fivos DOGANIS
> Cc : artoolkit@h ..................
> Objet : RE: camera calibration
>
> Hi,
>
> Hirokazu, could you please explain us how the two sets of parameters  
> relate to each other?
>
>
> Please look at:
> http://www.sys.im.hiroshima-cu.ac.jp/people/kato/etc/ART02/ART02- 
> Tutorial.pdf
>
> --
> ------------------------------------------------------------------
>   Hirokazu Kato
>   Faculty of Information Sciences
>   Hiroshima City University               Phone: +81-82-830-1705
>   Email: kato@s ........................     Fax: +81-82-830-1435
>   URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/
>

Fivos, what we are doing is calibrating the camera using the openCV matlab
implementation.  We are using the center of projection and focal length as
normal in the "camera parameter" matrix.  however, we set all distortion
factors to zero.  The distortion paramters are chiefly used in the
functions

observed2ideal and ideal2observed

since we are doing video dewarping externally in
our video library, we are assuming "perfectly straight" video in the
ARToolkit.  so to bypass all distortion corrections, we have just
effectively removed these two functions.  For simplicity we didn't
actually remove them, but we commented out everything and just return the
same values that are passed in.  this is a temporary solution of course.

To really integrate the openCV and/or Zhang calibration into the
ARToolkit, you would need to just rewrite these two functions,
observed2ideal and ideal2observed with the suitable mapping algorithms for
their distortion parameters (not the same as ARToolkits).  correct me if
im wrong Dr. Kato.

For the openCV version, like blair said, the observed2ideal function is
very simple, and the matlab calibration toolbox actually explicity gives
the math for dewarping points in its documentation.  However the inverse
function, ideal2observed, is non-trivial, and I believe the documentation
just gives references to some reseach papers that deal with this problem.

hope this helps,
brendan

----------------------------
|     Brendan Hannigan     |
|   brendan@c ............  |
----------------------------

On Mon, 3 Mar 2003, Fivos DOGANIS wrote:

> Hello Blair and Brendan,
>
> Of course, I could just bypass the whole camera transformation and just do
> my own, but I wanted to know if there was a way to do things by using the
> ARToolKit. I'd like to compare the tracking results of the toolkit, using an
> external calibration program (OpenCV) with the results I would get by doing
> my own transformation.
>
> As for the cost for dewarping, I don't think it is that expensive whan you
> are drawing on a texture.
>
> Thanks for the idea for setting the parameters to 0. I'll try that one and
> therefore only do my undistortion by hand, while using the proper values for
> the other camera parameters. Which parameters do you turn to 0? In the
> tutorial, I read:
>
> dist_factor[0] = x0
> dist_factor[1] = y0
> dist_factor[2] = 100 000 000 * f
> dist_factor[3] = s
>
> I guess you cannot turn the centre of the distortion and the general scaling
> factor to 0?
>
> Best regards,
>
> Fivos
>
> -----Message d'origine-----
> De : Blair MacIntyre [mailto:blair@c ............]
> Envoyé : lundi 3 mars 2003 17:38
> À : Fivos DOGANIS
> Cc : Hirokazu Kato; artoolkit@h ..................; Brendan Hannigan
> Objet : Re: camera calibration
>
>
> Fivos,
>
> I don't believe there is a straightforward conversion in the parameters.
>
> Talk to Brendan Hannigan (brendan@c ............), a student in my lab.
> He's been using the "better" calibration algorithms, and is in the
> process of more fully integrating them into the AR Toolkit;  the first
> pass was to use the OpenCV library (I believe) to dewarp the images
> before returning them to ARTK, and then set the ARTK distortion
> parameters to "0".  This has been working fine for us.
>
> He is planning on implementing "observedtoideal" for these distortion
> parameters (apparently it's straightforward) so we can avoid the cost
> of dewarping, but the reverse is not so simple.  He may be interested
> in getting some help (especially on that latter problem, the going from
> dewarped to warped space) ...
>
>
> On Monday, March 3, 2003, at 11:18  AM, Fivos DOGANIS wrote:
>
> > Dear Hirokazu,
> >  
> > Thank you for your reply, but I have already read the whole tutorial
> > over and over. I have tries to match the parameters in dist_factor
> > with the "classic" camera calibration parameters used by Matlab and
> > OpenCV. I was assuming that:
> >  
> > 1-f = k1
> >  
> > f being the distortion factor in ARToolKit and k1 being the first
> > coefficient of the radial distortion polynomial...
> > But my computations only match this equality up to a translation of
> > the centre of distortion.
> > Has anyone a clue about this?
> >  
> > Best regards,
> >  
> > Fivos
> >  
> >
> > -----Message d'origine-----
> > De : Hirokazu Kato [mailto:kato@s ........................]
> > Envoyé : lundi 3 mars 2003 16:59
> > À : Fivos DOGANIS
> > Cc : artoolkit@h ..................
> > Objet : RE: camera calibration
> >
> > Hi,
> >
> > Hirokazu, could you please explain us how the two sets of parameters
> > relate to each other?
> >
> >
> > Please look at:
> > http://www.sys.im.hiroshima-cu.ac.jp/people/kato/etc/ART02/ART02-
> > Tutorial.pdf
> >
> > --
> > ------------------------------------------------------------------
> >   Hirokazu Kato
> >   Faculty of Information Sciences
> >   Hiroshima City University               Phone: +81-82-830-1705
> >   Email: kato@s ........................     Fax: +81-82-830-1435
> >   URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/
> >
>
>

Fivos, what we are doing is calibrating the camera using the openCV matlab
implementation.  We are using the center of projection and focal length as
normal in the "camera parameter" matrix.  however, we set all distortion
factors to zero.  The distortion paramters are chiefly used in the
functions

observed2ideal and ideal2observed

since we are doing video dewarping externally in
our video library, we are assuming "perfectly straight" video in the
ARToolkit.  so to bypass all distortion corrections, we have just
effectively removed these two functions.  For simplicity we didn't
actually remove them, but we commented out everything and just return the
same values that are passed in.  this is a temporary solution of course.

To really integrate the openCV and/or Zhang calibration into the
ARToolkit, you would need to just rewrite these two functions,
observed2ideal and ideal2observed with the suitable mapping algorithms for
their distortion parameters (not the same as ARToolkits).  correct me if
im wrong Dr. Kato.

For the openCV version, like blair said, the observed2ideal function is
very simple, and the matlab calibration toolbox actually explicity gives
the math for dewarping points in its documentation.  However the inverse
function, ideal2observed, is non-trivial, and I believe the documentation
just gives references to some reseach papers that deal with this problem.

hope this helps,
brendan

----------------------------
|     Brendan Hannigan     |
|   brendan@c ............  |
----------------------------

On Mon, 3 Mar 2003, Fivos DOGANIS wrote:

> Hello Blair and Brendan,
>
> Of course, I could just bypass the whole camera transformation and just do
> my own, but I wanted to know if there was a way to do things by using the
> ARToolKit. I'd like to compare the tracking results of the toolkit, using an
> external calibration program (OpenCV) with the results I would get by doing
> my own transformation.
>
> As for the cost for dewarping, I don't think it is that expensive whan you
> are drawing on a texture.
>
> Thanks for the idea for setting the parameters to 0. I'll try that one and
> therefore only do my undistortion by hand, while using the proper values for
> the other camera parameters. Which parameters do you turn to 0? In the
> tutorial, I read:
>
> dist_factor[0] = x0
> dist_factor[1] = y0
> dist_factor[2] = 100 000 000 * f
> dist_factor[3] = s
>
> I guess you cannot turn the centre of the distortion and the general scaling
> factor to 0?
>
> Best regards,
>
> Fivos
>
> -----Message d'origine-----
> De : Blair MacIntyre [mailto:blair@c ............]
> Envoyé : lundi 3 mars 2003 17:38
> À : Fivos DOGANIS
> Cc : Hirokazu Kato; artoolkit@h ..................; Brendan Hannigan
> Objet : Re: camera calibration
>
>
> Fivos,
>
> I don't believe there is a straightforward conversion in the parameters.
>
> Talk to Brendan Hannigan (brendan@c ............), a student in my lab.
> He's been using the "better" calibration algorithms, and is in the
> process of more fully integrating them into the AR Toolkit;  the first
> pass was to use the OpenCV library (I believe) to dewarp the images
> before returning them to ARTK, and then set the ARTK distortion
> parameters to "0".  This has been working fine for us.
>
> He is planning on implementing "observedtoideal" for these distortion
> parameters (apparently it's straightforward) so we can avoid the cost
> of dewarping, but the reverse is not so simple.  He may be interested
> in getting some help (especially on that latter problem, the going from
> dewarped to warped space) ...
>
>
> On Monday, March 3, 2003, at 11:18  AM, Fivos DOGANIS wrote:
>
> > Dear Hirokazu,
> >  
> > Thank you for your reply, but I have already read the whole tutorial
> > over and over. I have tries to match the parameters in dist_factor
> > with the "classic" camera calibration parameters used by Matlab and
> > OpenCV. I was assuming that:
> >  
> > 1-f = k1
> >  
> > f being the distortion factor in ARToolKit and k1 being the first
> > coefficient of the radial distortion polynomial...
> > But my computations only match this equality up to a translation of
> > the centre of distortion.
> > Has anyone a clue about this?
> >  
> > Best regards,
> >  
> > Fivos
> >  
> >
> > -----Message d'origine-----
> > De : Hirokazu Kato [mailto:kato@s ........................]
> > Envoyé : lundi 3 mars 2003 16:59
> > À : Fivos DOGANIS
> > Cc : artoolkit@h ..................
> > Objet : RE: camera calibration
> >
> > Hi,
> >
> > Hirokazu, could you please explain us how the two sets of parameters
> > relate to each other?
> >
> >
> > Please look at:
> > http://www.sys.im.hiroshima-cu.ac.jp/people/kato/etc/ART02/ART02-
> > Tutorial.pdf
> >
> > --
> > ------------------------------------------------------------------
> >   Hirokazu Kato
> >   Faculty of Information Sciences
> >   Hiroshima City University               Phone: +81-82-830-1705
> >   Email: kato@s ........................     Fax: +81-82-830-1435
> >   URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/
> >
>
>

Dear Hirokazu,
 
Thank you for your reply, but I have already read the whole tutorial over and over. I have tries to match the parameters in dist_factor with the "classic" camera calibration parameters used by Matlab and OpenCV. I was assuming that:
 
1-f = k1
 
f being the distortion factor in ARToolKit and k1 being the first coefficient of the radial distortion polynomial...
But my computations only match this equality up to a translation of the centre of distortion.
Has anyone a clue about this?
 
Best regards,
 
Fivos
 

-----Message d'origine-----
De : Hirokazu Kato [mailto:kato@s ........................]
Envoyé : lundi 3 mars 2003 16:59
À : Fivos DOGANIS
Cc : artoolkit@h ..................
Objet : RE: camera calibration


Hi,


Hirokazu, could you please explain us how the two sets of parameters relate to each other?


Please look at:
http://www.sys.im.hiroshima-cu.ac.jp/people/kato/etc/ART02/ART02-Tutorial.pdf

-- 
------------------------------------------------------------------
  Hirokazu Kato
  Faculty of Information Sciences
  Hiroshima City University               Phone: +81-82-830-1705
  Email: kato@s ........................     Fax: +81-82-830-1435
  URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/

Hello Blair and Brendan,

Of course, I could just bypass the whole camera transformation and just do
my own, but I wanted to know if there was a way to do things by using the
ARToolKit. I'd like to compare the tracking results of the toolkit, using an
external calibration program (OpenCV) with the results I would get by doing
my own transformation.

As for the cost for dewarping, I don't think it is that expensive whan you
are drawing on a texture.

Thanks for the idea for setting the parameters to 0. I'll try that one and
therefore only do my undistortion by hand, while using the proper values for
the other camera parameters. Which parameters do you turn to 0? In the
tutorial, I read:

dist_factor[0] = x0
dist_factor[1] = y0
dist_factor[2] = 100 000 000 * f
dist_factor[3] = s

I guess you cannot turn the centre of the distortion and the general scaling
factor to 0?

Best regards,

Fivos

-----Message d'origine-----
De : Blair MacIntyre [mailto:blair@c ............]
Envoyé : lundi 3 mars 2003 17:38
À : Fivos DOGANIS
Cc : Hirokazu Kato; artoolkit@h ..................; Brendan Hannigan
Objet : Re: camera calibration


Fivos,

I don't believe there is a straightforward conversion in the parameters.

Talk to Brendan Hannigan (brendan@c ............), a student in my lab.
He's been using the "better" calibration algorithms, and is in the
process of more fully integrating them into the AR Toolkit;  the first
pass was to use the OpenCV library (I believe) to dewarp the images
before returning them to ARTK, and then set the ARTK distortion
parameters to "0".  This has been working fine for us.

He is planning on implementing "observedtoideal" for these distortion
parameters (apparently it's straightforward) so we can avoid the cost
of dewarping, but the reverse is not so simple.  He may be interested
in getting some help (especially on that latter problem, the going from
dewarped to warped space) ...


On Monday, March 3, 2003, at 11:18  AM, Fivos DOGANIS wrote:

> Dear Hirokazu,
>  
> Thank you for your reply, but I have already read the whole tutorial
> over and over. I have tries to match the parameters in dist_factor
> with the "classic" camera calibration parameters used by Matlab and
> OpenCV. I was assuming that:
>  
> 1-f = k1
>  
> f being the distortion factor in ARToolKit and k1 being the first
> coefficient of the radial distortion polynomial...
> But my computations only match this equality up to a translation of
> the centre of distortion.
> Has anyone a clue about this?
>  
> Best regards,
>  
> Fivos
>  
>
> -----Message d'origine-----
> De : Hirokazu Kato [mailto:kato@s ........................]
> Envoyé : lundi 3 mars 2003 16:59
> À : Fivos DOGANIS
> Cc : artoolkit@h ..................
> Objet : RE: camera calibration
>
> Hi,
>
> Hirokazu, could you please explain us how the two sets of parameters
> relate to each other?
>
>
> Please look at:
> http://www.sys.im.hiroshima-cu.ac.jp/people/kato/etc/ART02/ART02-
> Tutorial.pdf
>
> --
> ------------------------------------------------------------------
>   Hirokazu Kato
>   Faculty of Information Sciences
>   Hiroshima City University               Phone: +81-82-830-1705
>   Email: kato@s ........................     Fax: +81-82-830-1435
>   URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/
>

Hello Blair and Brendan,

Of course, I could just bypass the whole camera transformation and just do
my own, but I wanted to know if there was a way to do things by using the
ARToolKit. I'd like to compare the tracking results of the toolkit, using an
external calibration program (OpenCV) with the results I would get by doing
my own transformation.

As for the cost for dewarping, I don't think it is that expensive whan you
are drawing on a texture.

Thanks for the idea for setting the parameters to 0. I'll try that one and
therefore only do my undistortion by hand, while using the proper values for
the other camera parameters. Which parameters do you turn to 0? In the
tutorial, I read:

dist_factor[0] = x0
dist_factor[1] = y0
dist_factor[2] = 100 000 000 * f
dist_factor[3] = s

I guess you cannot turn the centre of the distortion and the general scaling
factor to 0?

Best regards,

Fivos

-----Message d'origine-----
De : Blair MacIntyre [mailto:blair@c ............]
Envoyé : lundi 3 mars 2003 17:38
À : Fivos DOGANIS
Cc : Hirokazu Kato; artoolkit@h ..................; Brendan Hannigan
Objet : Re: camera calibration


Fivos,

I don't believe there is a straightforward conversion in the parameters.

Talk to Brendan Hannigan (brendan@c ............), a student in my lab.
He's been using the "better" calibration algorithms, and is in the
process of more fully integrating them into the AR Toolkit;  the first
pass was to use the OpenCV library (I believe) to dewarp the images
before returning them to ARTK, and then set the ARTK distortion
parameters to "0".  This has been working fine for us.

He is planning on implementing "observedtoideal" for these distortion
parameters (apparently it's straightforward) so we can avoid the cost
of dewarping, but the reverse is not so simple.  He may be interested
in getting some help (especially on that latter problem, the going from
dewarped to warped space) ...


On Monday, March 3, 2003, at 11:18  AM, Fivos DOGANIS wrote:

> Dear Hirokazu,
>  
> Thank you for your reply, but I have already read the whole tutorial
> over and over. I have tries to match the parameters in dist_factor
> with the "classic" camera calibration parameters used by Matlab and
> OpenCV. I was assuming that:
>  
> 1-f = k1
>  
> f being the distortion factor in ARToolKit and k1 being the first
> coefficient of the radial distortion polynomial...
> But my computations only match this equality up to a translation of
> the centre of distortion.
> Has anyone a clue about this?
>  
> Best regards,
>  
> Fivos
>  
>
> -----Message d'origine-----
> De : Hirokazu Kato [mailto:kato@s ........................]
> Envoyé : lundi 3 mars 2003 16:59
> À : Fivos DOGANIS
> Cc : artoolkit@h ..................
> Objet : RE: camera calibration
>
> Hi,
>
> Hirokazu, could you please explain us how the two sets of parameters
> relate to each other?
>
>
> Please look at:
> http://www.sys.im.hiroshima-cu.ac.jp/people/kato/etc/ART02/ART02-
> Tutorial.pdf
>
> --
> ------------------------------------------------------------------
>   Hirokazu Kato
>   Faculty of Information Sciences
>   Hiroshima City University               Phone: +81-82-830-1705
>   Email: kato@s ........................     Fax: +81-82-830-1435
>   URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/
>

Brendan,

Thank you for your detailed reply, I'll investigate on these topics (when I
have some time!).
Do not hesitate to ask me for anything, just in case I can answer :)

Regards,

Fivos

-----Message d'origine-----
De : Brendan [mailto:brendan@c ............]
Envoyé : lundi 3 mars 2003 18:16
À : Fivos DOGANIS
Cc : artoolkit@h ..................; Blair MacIntyre
Objet : RE: camera calibration



Fivos, what we are doing is calibrating the camera using the openCV matlab
implementation.  We are using the center of projection and focal length as
normal in the "camera parameter" matrix.  however, we set all distortion
factors to zero.  The distortion paramters are chiefly used in the
functions

observed2ideal and ideal2observed

since we are doing video dewarping externally in
our video library, we are assuming "perfectly straight" video in the
ARToolkit.  so to bypass all distortion corrections, we have just
effectively removed these two functions.  For simplicity we didn't
actually remove them, but we commented out everything and just return the
same values that are passed in.  this is a temporary solution of course.

To really integrate the openCV and/or Zhang calibration into the
ARToolkit, you would need to just rewrite these two functions,
observed2ideal and ideal2observed with the suitable mapping algorithms for
their distortion parameters (not the same as ARToolkits).  correct me if
im wrong Dr. Kato.

For the openCV version, like blair said, the observed2ideal function is
very simple, and the matlab calibration toolbox actually explicity gives
the math for dewarping points in its documentation.  However the inverse
function, ideal2observed, is non-trivial, and I believe the documentation
just gives references to some reseach papers that deal with this problem.

hope this helps,
brendan

----------------------------
|     Brendan Hannigan     |
|   brendan@c ............  |
----------------------------

On Mon, 3 Mar 2003, Fivos DOGANIS wrote:

> Hello Blair and Brendan,
>
> Of course, I could just bypass the whole camera transformation and just do
> my own, but I wanted to know if there was a way to do things by using the
> ARToolKit. I'd like to compare the tracking results of the toolkit, using
an
> external calibration program (OpenCV) with the results I would get by
doing
> my own transformation.
>
> As for the cost for dewarping, I don't think it is that expensive whan you
> are drawing on a texture.
>
> Thanks for the idea for setting the parameters to 0. I'll try that one and
> therefore only do my undistortion by hand, while using the proper values
for
> the other camera parameters. Which parameters do you turn to 0? In the
> tutorial, I read:
>
> dist_factor[0] = x0
> dist_factor[1] = y0
> dist_factor[2] = 100 000 000 * f
> dist_factor[3] = s
>
> I guess you cannot turn the centre of the distortion and the general
scaling
> factor to 0?
>
> Best regards,
>
> Fivos
>
> -----Message d'origine-----
> De : Blair MacIntyre [mailto:blair@c ............]
> Envoyé : lundi 3 mars 2003 17:38
> À : Fivos DOGANIS
> Cc : Hirokazu Kato; artoolkit@h ..................; Brendan Hannigan
> Objet : Re: camera calibration
>
>
> Fivos,
>
> I don't believe there is a straightforward conversion in the parameters.
>
> Talk to Brendan Hannigan (brendan@c ............), a student in my lab.
> He's been using the "better" calibration algorithms, and is in the
> process of more fully integrating them into the AR Toolkit;  the first
> pass was to use the OpenCV library (I believe) to dewarp the images
> before returning them to ARTK, and then set the ARTK distortion
> parameters to "0".  This has been working fine for us.
>
> He is planning on implementing "observedtoideal" for these distortion
> parameters (apparently it's straightforward) so we can avoid the cost
> of dewarping, but the reverse is not so simple.  He may be interested
> in getting some help (especially on that latter problem, the going from
> dewarped to warped space) ...
>
>
> On Monday, March 3, 2003, at 11:18  AM, Fivos DOGANIS wrote:
>
> > Dear Hirokazu,
> >  
> > Thank you for your reply, but I have already read the whole tutorial
> > over and over. I have tries to match the parameters in dist_factor
> > with the "classic" camera calibration parameters used by Matlab and
> > OpenCV. I was assuming that:
> >  
> > 1-f = k1
> >  
> > f being the distortion factor in ARToolKit and k1 being the first
> > coefficient of the radial distortion polynomial...
> > But my computations only match this equality up to a translation of
> > the centre of distortion.
> > Has anyone a clue about this?
> >  
> > Best regards,
> >  
> > Fivos
> >  
> >
> > -----Message d'origine-----
> > De : Hirokazu Kato [mailto:kato@s ........................]
> > Envoyé : lundi 3 mars 2003 16:59
> > À : Fivos DOGANIS
> > Cc : artoolkit@h ..................
> > Objet : RE: camera calibration
> >
> > Hi,
> >
> > Hirokazu, could you please explain us how the two sets of parameters
> > relate to each other?
> >
> >
> > Please look at:
> > http://www.sys.im.hiroshima-cu.ac.jp/people/kato/etc/ART02/ART02-
> > Tutorial.pdf
> >
> > --
> > ------------------------------------------------------------------
> >   Hirokazu Kato
> >   Faculty of Information Sciences
> >   Hiroshima City University               Phone: +81-82-830-1705
> >   Email: kato@s ........................     Fax: +81-82-830-1435
> >   URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/
> >
>
>

Brendan,

Thank you for your detailed reply, I'll investigate on these topics (when I
have some time!).
Do not hesitate to ask me for anything, just in case I can answer :)

Regards,

Fivos

-----Message d'origine-----
De : Brendan [mailto:brendan@c ............]
Envoyé : lundi 3 mars 2003 18:16
À : Fivos DOGANIS
Cc : artoolkit@h ..................; Blair MacIntyre
Objet : RE: camera calibration



Fivos, what we are doing is calibrating the camera using the openCV matlab
implementation.  We are using the center of projection and focal length as
normal in the "camera parameter" matrix.  however, we set all distortion
factors to zero.  The distortion paramters are chiefly used in the
functions

observed2ideal and ideal2observed

since we are doing video dewarping externally in
our video library, we are assuming "perfectly straight" video in the
ARToolkit.  so to bypass all distortion corrections, we have just
effectively removed these two functions.  For simplicity we didn't
actually remove them, but we commented out everything and just return the
same values that are passed in.  this is a temporary solution of course.

To really integrate the openCV and/or Zhang calibration into the
ARToolkit, you would need to just rewrite these two functions,
observed2ideal and ideal2observed with the suitable mapping algorithms for
their distortion parameters (not the same as ARToolkits).  correct me if
im wrong Dr. Kato.

For the openCV version, like blair said, the observed2ideal function is
very simple, and the matlab calibration toolbox actually explicity gives
the math for dewarping points in its documentation.  However the inverse
function, ideal2observed, is non-trivial, and I believe the documentation
just gives references to some reseach papers that deal with this problem.

hope this helps,
brendan

----------------------------
|     Brendan Hannigan     |
|   brendan@c ............  |
----------------------------

On Mon, 3 Mar 2003, Fivos DOGANIS wrote:

> Hello Blair and Brendan,
>
> Of course, I could just bypass the whole camera transformation and just do
> my own, but I wanted to know if there was a way to do things by using the
> ARToolKit. I'd like to compare the tracking results of the toolkit, using
an
> external calibration program (OpenCV) with the results I would get by
doing
> my own transformation.
>
> As for the cost for dewarping, I don't think it is that expensive whan you
> are drawing on a texture.
>
> Thanks for the idea for setting the parameters to 0. I'll try that one and
> therefore only do my undistortion by hand, while using the proper values
for
> the other camera parameters. Which parameters do you turn to 0? In the
> tutorial, I read:
>
> dist_factor[0] = x0
> dist_factor[1] = y0
> dist_factor[2] = 100 000 000 * f
> dist_factor[3] = s
>
> I guess you cannot turn the centre of the distortion and the general
scaling
> factor to 0?
>
> Best regards,
>
> Fivos
>
> -----Message d'origine-----
> De : Blair MacIntyre [mailto:blair@c ............]
> Envoyé : lundi 3 mars 2003 17:38
> À : Fivos DOGANIS
> Cc : Hirokazu Kato; artoolkit@h ..................; Brendan Hannigan
> Objet : Re: camera calibration
>
>
> Fivos,
>
> I don't believe there is a straightforward conversion in the parameters.
>
> Talk to Brendan Hannigan (brendan@c ............), a student in my lab.
> He's been using the "better" calibration algorithms, and is in the
> process of more fully integrating them into the AR Toolkit;  the first
> pass was to use the OpenCV library (I believe) to dewarp the images
> before returning them to ARTK, and then set the ARTK distortion
> parameters to "0".  This has been working fine for us.
>
> He is planning on implementing "observedtoideal" for these distortion
> parameters (apparently it's straightforward) so we can avoid the cost
> of dewarping, but the reverse is not so simple.  He may be interested
> in getting some help (especially on that latter problem, the going from
> dewarped to warped space) ...
>
>
> On Monday, March 3, 2003, at 11:18  AM, Fivos DOGANIS wrote:
>
> > Dear Hirokazu,
> >  
> > Thank you for your reply, but I have already read the whole tutorial
> > over and over. I have tries to match the parameters in dist_factor
> > with the "classic" camera calibration parameters used by Matlab and
> > OpenCV. I was assuming that:
> >  
> > 1-f = k1
> >  
> > f being the distortion factor in ARToolKit and k1 being the first
> > coefficient of the radial distortion polynomial...
> > But my computations only match this equality up to a translation of
> > the centre of distortion.
> > Has anyone a clue about this?
> >  
> > Best regards,
> >  
> > Fivos
> >  
> >
> > -----Message d'origine-----
> > De : Hirokazu Kato [mailto:kato@s ........................]
> > Envoyé : lundi 3 mars 2003 16:59
> > À : Fivos DOGANIS
> > Cc : artoolkit@h ..................
> > Objet : RE: camera calibration
> >
> > Hi,
> >
> > Hirokazu, could you please explain us how the two sets of parameters
> > relate to each other?
> >
> >
> > Please look at:
> > http://www.sys.im.hiroshima-cu.ac.jp/people/kato/etc/ART02/ART02-
> > Tutorial.pdf
> >
> > --
> > ------------------------------------------------------------------
> >   Hirokazu Kato
> >   Faculty of Information Sciences
> >   Hiroshima City University               Phone: +81-82-830-1705
> >   Email: kato@s ........................     Fax: +81-82-830-1435
> >   URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/
> >
>
>

Hi,


Hirokazu, could you please explain us how the two sets of parameters relate to each other?


Please look at:
http://www.sys.im.hiroshima-cu.ac.jp/people/kato/etc/ART02/ART02-Tutorial.pdf

-- 
------------------------------------------------------------------
  Hirokazu Kato
  Faculty of Information Sciences
  Hiroshima City University               Phone: +81-82-830-1705
  Email: kato@s ........................     Fax: +81-82-830-1435
  URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/

>To really integrate the openCV and/or Zhang calibration into the
>ARToolkit, you would need to just rewrite these two functions,
>observed2ideal and ideal2observed with the suitable mapping algorithms for
>their distortion parameters (not the same as ARToolkits).  correct me if
>im wrong Dr. Kato.

You are right!!
Thank you for explanations.


-- 
------------------------------------------------------------------
  Hirokazu Kato
  Faculty of Information Sciences
  Hiroshima City University               Phone: +81-82-830-1705
  Email: kato@s ........................     Fax: +81-82-830-1435
  URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/