View Full Version : Replies of MDLT question

07-08-2003, 05:23 PM
Here are the replies of MDLT Question.Thank you all! And specially I thanks
Tomislav Pribanic.
The Original question is:

----- Original Message -----
From: "严开涛"
Sent: Tuesday, May 27, 2003 2:09 PM
Subject: [BIOMCH-L] MDLT question

Now I use modified DLT method(10 parameters) to improve the accuracy.But
sometime I can't get a stable set of camera calibration data.Are there
anybody meet the same trouble is using MDLT? And are there any method to
resolve this problem?
I would be glad to post summary of replies to the ListServer.

Sincerly Kaitao Yan
DorealSoft Company

Here are the replies:

----- Original Message -----
From: "Tomislav Pribanic"
To: "严开涛"
Sent: Wednesday, May 28, 2003 12:00 AM
Subject: Re: [BIOMCH-L] MDLT question

> Dear Kaiato,
> convergence of MDLT algorithm highly depends of suitability of initial set
> of camera parameters solution. MDLT does have its theoretical
> and reasons due to fact that linearization of camera model equations lead
> one more additional camera parameter. However, the way how you eliminate
> redundant parameter also carry its weight. A while ago I put the version
> MDLT algorithm on ISB software page and I think it is still there. You can
> try to use that version and see if it converges. Inside the matlab routine
> there is an explanation how function works. Nevertheless, it has been
> sometime since I wrote it and today I would recommend some other things.
> First at all, DLT algorithm and/or MDLT in essence depends on 3D
> cage/structure. Although it is very effective approach for camera
> calibration, nowadays trend is to create more user-friendly 3D kinematic
> systems, thus avoiding manipulating and utilizing with sometimes
> cage. Therefore, other approaches for camera calibration have been and are
> developing such as plane and wand calibration. These approaches are
> computationally more complex, but offer greater flexibility particularly
> outdoor measurements where the luxury of one-time laboratory calibration
> some period of time (set of measurements) is excluded.
> I do not know the purpose of your work, but unless you really strive for
> highest degree of reconstruction accuracy (calibration cage is still at
> least a bit more accurate then plane or wand calibration regardless what
> some commercial suppliers might tell you) you can switch to plane or wand
> calibration. If for some reasons you need to stick with 3D calibration
> then MDLT by itself, even with convergent set of solution, may not do much
> difference. Namely, some investigators stated that MDLT makes sense only
> when one is extrapolating outside calibration volume. For most
> one can control his subjects to remain inside the calibration volume and
> not there is panning camera system as alternative.
> One thing what would make a difference (if we exclude some extraordinary
> circumstances), regardless of type of calibration is compensation for
> non-linear camera lens distortion.
> In brief, the 'golden algorithm' for use of 3D calibration cage would be:
> 1. Carry out DLT calibration
> 2. extract camera parameters with so-called qr decomposition of camera
> matrix (there is a paper explaining qr decomposition for the purpose of
> camera calibration, but I cannot think of it now; if you fail finding it
> also I'll try look for it)
> 3. Include non-linear distortion parameters and along with camera
> from previous step start non-linear minimization procedure (I recommend
> Levenberg-Marquardt algorithm, but there are many others as well) and
> your camera parameters through certain number of iterations until
> set of camera parameters solution is reached.
> Regards, Tomislav.
> Tomislav Pribanic, M.Sc., EE
> Department for Electronic Systems and Information Processing
> Faculty of Electrical Engineering and Computing
> 3 Unska, 10000 Zagreb, Croatia
> tel. ..385 1 612 98 67, fax. ..385 1 612 96 52
> E-mail : tomislav.pribanic@fer.hr

----- Original Message -----
From: "Kjartan Halvorsen"
To: "严开涛"
Sent: Wednesday, May 28, 2003 3:09 PM
Subject: Re: [BIOMCH-L] MDLT question

Dear mr Kaitao Yan,

It is important for numerical stability that the coordinates, both 2D and
are scaled so that they are on average of the order of 1.
See R. Hartley: "In defense of the eight-point algorithm" IEEE Tr PAMI
vol. 19 (6), 1997

Maybe this is the problem with your computations.

Yours sincerely,

Kjartan Halvorsen

To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl
For information and archives: http://isb.ri.ccf.org/biomch-l