Hi,

I'm currently building an application with ARToolKitPlus (I've actually
built a java bridge with JNI: if you're interested I could put it online
once its a bit more tested). I'm trying to detect very small markers and
have a few questions.

1) What is the smallest marker size (in centimeters) (BCH or simple) I
should be able to detect with a 320x240 camera? Does the actual (real)
marker size affects recognition?
2) I have noticed that no matter what the actual size of the markers, I
have trouble detecting them when they occupy a small portion of the image.
I seem to be only able to detect markers when they are close enough to the
camera to occupy 20% of the image height. What are the parameters that I
could play with to adjust this?

Thanks,

J. S. Senecal

Hi,

As Wayne already pointed out, is is important how large the marker shows
up in your image (in pixels, not in percent!), rather than how large it
really is. Here are some factors that decide whether or not ARTK/ARTK+
will be able to detect a marker:
o absolute size in pixels
  should be at least roughly 15 pixels in each direction so that
  the unprojection step can extract the original marker image
o orientation
  strong tilt angles will make the marker thinner
  (see my previous point)
o noise in the camera image
  if the image is too noisy detection quality will
  drop considerably
o quality of camera calibration
  people often forget calibrate to their camera and
  just use calibration a file that come with ARTK.
o speed of motion
  fast camera movement tends to blur the image
o field of view of the camera
  wide FOV camera ("fish-eye") make it harder for
  ARTK to undistort the image
o resolution of the camera
  camera with higher resolutions will naturally be
  able to detect smaller markers
o quality of the camera lens
  people tend to forget that a tiny (<1mm) plastic lens
  as often used in chep webcams will not be able to
  project an image accurately onto the CCD.

In case performance is important: I noticed that a correct camera
calibration can make a huge (up to 100% on slow devices) difference in
the speed of pose estimation as the estimator will converge faster with
more accurate input!

320x240 is really a very low camera resolution. PCs have no problem
today working with 640x480 at high frame rates.

hope that helps,
bye,
Daniel

Hi,

The marker size is determined by the number of pixels that are used  
to capture the marker. For example, a 10 cm sized marker at 1 metre  
will be visible in the same way as a 20 cm sized marker at 2 metres.

So in order to answer your question you need to know roughly what the  
working volume will be. Then you could do something like putting  
various sized markers at the maximum distance and evaluate the  
performance. Each camera is different so this is why you need to test  
it.

regards,
Wayne

------------------------------------------------------------------------ 
-
Dr Wayne Piekarski                            pho:         
+61-8-8302-5070
Researcher and Senior Lecturer                fax:         
+61-8-8302-3381
Co Director - Wearable Computer Lab           mob:         
+61-407-395-889
School of Computer and Information Science    ema:   
wayne@c ..............
University of South Australia                 www: http:// 
www.tinmith.net


On 12/11/2006, at 1:27 PM, js@d ....... wrote:

> Hi,
>
> I'm currently building an application with ARToolKitPlus (I've  
> actually
> built a java bridge with JNI: if you're interested I could put it  
> online
> once its a bit more tested). I'm trying to detect very small  
> markers and
> have a few questions.
>
> 1) What is the smallest marker size (in centimeters) (BCH or simple) I
> should be able to detect with a 320x240 camera? Does the actual (real)
> marker size affects recognition?
> 2) I have noticed that no matter what the actual size of the  
> markers, I
> have trouble detecting them when they occupy a small portion of the  
> image.
> I seem to be only able to detect markers when they are close enough  
> to the
> camera to occupy 20% of the image height. What are the parameters  
> that I
> could play with to adjust this?
>
> Thanks,
>
> J. S. Senecal
>
>
>