Hi!

I've got some questions that might have been answered before, but I hope
someone can help me :-)

First of all, thanks Kato, for your quick respond to my previous questions!

My questions:
1) I am using a Sony CCD-TR2000E with Matrox MeteorII capture/video card on 
a Windows NT system. ARToolkit is 2.52 with VisSDK 1.2. The camera has 
support for up to 768x576 pixels, but it seems like the default in ARToolkit 
is 320x240. On another computer, using a webcam, I was able to change image 
dimensions by running excamera.exe, but with the Matrox MeteorII, I do not 
have the option to press the "F"-button in excamera.exe. I have recompiled 
excamera.exe that follows VisSDK, but this does not change anything. So 
finally to my question:
   Is there another way to change the image dimensions/resolution?
Maybe by setting dimensions in ARToolkit, running a
program, editing the registry-value in Windows, or something like that? I 
would like to run on maximum resolution.

2) When setting a zoom factor in argInit(), are the image recognition
algorithms performed on the zoomed image, or on the original one? I guess on 
the original, but I would like to get it verified.

3) Is the translation in the patt_trans[3][4] given in mm? And the rotation 
calculated in radians?

4) If a marker is far away from the camera, on some frames ARToolkit cannot 
detect the marker. Is there a way to alert when it is not found (apart from 
the fact that you can see some "glitches")? How do I know when ARToolkit has 
found a specific marker? Is it possible to make a printf() for each frame 
that says :"hiro.patt found" or "hiro.patt not found" - based on ARToolkit's 
own criteria?

5) I guess that ARToolkit recognises the squares around the markers more
easily than it can find the patterns inside the squares (separating the 
different marker-patterns). So, even if ARToolkit detects a marker, it might 
not display graphics on top of it, because no matching template (of the 
recorded markers) are found. Is this correct?

6) What if ARToolkit misses tracking in some frames - does it use knowledge 
about the marker's position from previous frames? In case, how many previous 
frames are used? What are the criteria for using previous frames? Does the 
marker's confidence value have to be below a certain level before it is 
acceptable to use knowledge about previous positions, or the other way 
around?

7) How is the confidence value calculated? Is it based on size/contrast in 
image/marker slant/lightning conditions, or others?

8) After running calib_camera2.exe on the same camera in different
lightning conditions, the camera parameters calculated in the two 
calibrations turned out to be quite different. Why is this the case? Are 
there certain criteria on how to run the calibration? At which distance from 
the camera should the images for the calibration be captured; close to the 
camera, far away from the camera, or both? How much slant on the markers 
should be present (much,little or both)? Should the images present very 
different distances or angles, or many similar cases?

Ok, that was a lot of questions! Hope somebody will take their time to
answer them :-)

Have a nice Easter !!

Best Regards,
Petter Lundevann

PS! It would be a great idea to create an archive for this mailinglist. I 
guess many of my questions have been asked and answered before, giving the 
developers and people on the list more questions to answer than necessary! 
Does anybody have a "private archive"? Have you saved this mailinglist for a 
period in time? It would be great to have this available to new users!


_________________________________________________________________
Join the world’s largest e-mail service with MSN Hotmail. 
http://www.hotmail.com

Hi,

>1) 
>   Is there another way to change the image dimensions/resolution?
>Maybe by setting dimensions in ARToolkit, running a
>program, editing the registry-value in Windows, or something like that? I 
>would like to run on maximum resolution.

Sorry. I don't know, because I don't use windows version.


>2) When setting a zoom factor in argInit(), are the image recognition
>algorithms performed on the zoomed image, or on the original one?

original one.


>3) Is the translation in the patt_trans[3][4] given in mm? And the rotation 
>calculated in radians?

Actually it depends on the calibration. While calibration, we have to specify
some information. If we input length in 'mm' for such information,
translation as a tracking result is represented in 'mm'.
If we input in 'inch', translation is represented in 'inch'.
Default is 'mm'.


>4) If a marker is far away from the camera, on some frames ARToolkit cannot 
>detect the marker. Is there a way to alert when it is not found (apart from 
>the fact that you can see some "glitches")? How do I know when ARToolkit has 
>found a specific marker? Is it possible to make a printf() for each frame 
>that says :"hiro.patt found" or "hiro.patt not found" - based on ARToolkit's 
>own criteria?

Please look at 'simpleTest.c'.

    /* check for object visibility */
    k = -1;
    for( j = 0; j < marker_num; j++ ) {
        if( patt_id == marker_info[j].id ) {
            if( k == -1 ) k = j;
            else if( marker_info[k].cf < marker_info[j].cf ) k = j;
        }
    }
    if( k == -1 ) {
        argSwapBuffers();
        return;
    }

You can change this code. This code looks for a marker with max cf value
for 'patt_id'. And 'k == -1' means that the marker was not found.


>5) I guess that ARToolkit recognises the squares around the markers more
>easily than it can find the patterns inside the squares (separating the 
>different marker-patterns). So, even if ARToolkit detects a marker, it might 
>not display graphics on top of it, because no matching template (of the 
>recorded markers) are found. Is this correct?

Yes. But arDetectMarker() uses tracking history. Even if template matching
was not done successfully, if there is a tracked marker with almost same size
and same position in previous frame, the marker is regarded as a same marker
of previous frame.
arDetectMarkerLite() do not use tracking history.
So if you want to use your own algorithm for this purpose, you should use
arDetectMarkerLite().


>6) What if ARToolkit misses tracking in some frames - does it use knowledge 
>about the marker's position from previous frames?

Yes, in case of arDetectMarker(). No, in case of arDetectMarkerLite().

> In case, how many previous 
>frames are used? 

Probably 3.


>What are the criteria for using previous frames? Does the 
>marker's confidence value have to be below a certain level before it is 
>acceptable to use knowledge about previous positions, or the other way 
>around?

The algorithm in arDetectMarker() is not so clear.
Please look at the code.


>7) How is the confidence value calculated? Is it based on size/contrast in 
>image/marker slant/lightning conditions, or others?

It is from template matching based on normalized correlation.


>8) After running calib_camera2.exe on the same camera in different
>lightning conditions, the camera parameters calculated in the two 
>calibrations turned out to be quite different. Why is this the case? Are 
>there certain criteria on how to run the calibration? At which distance from 
>the camera should the images for the calibration be captured; close to the 
>camera, far away from the camera, or both? How much slant on the markers 
>should be present (much,little or both)? Should the images present very 
>different distances or angles, or many similar cases?

This is very difficult to explain for me.
I wish someone who knows calibration process could answer about that.


-- 
------------------------------------------------------------------
  Hirokazu Kato
  Faculty of Information Sciences
  Hiroshima City University               Phone: 082-830-1705
  Email: kato@s ........................     Fax: 082-830-1435
  URL: http://www.sys.im.hiroshima-cu.ac.jp/people/kato/