AForge.NET

  :: AForge.NET Framework :: Articles :: Forums ::

Quadrilateral transformation distance estimation

Forum to discuss Glyph Recognition And Tracking Framework project, its features, applications, etc.

Quadrilateral transformation distance estimation

Postby moshe » Sat Feb 05, 2011 9:50 am

Hello all and thank you for the great tools you are developing in this project.

I'm working on a robotics project, and my intent is to use glyphs as reference points for the navigation system.
In order to do so, I need to extract the rotation and translation matrices derived from the quadrilateral transformation.
from what I see, there is no direct access to the matrices, only the multiplication results are given.
Is there a way to access the matrices, and, if anyone had tried it before, what are the accuracy levels I can expect?

Best regards and thanks for reading,

Moshe.
moshe
 
Posts: 2
Joined: Sat Feb 05, 2011 9:42 am

Re: Quadrilateral transformation distance estimation

Postby andrew.kirillov » Sat Feb 05, 2011 11:37 am

Hello,

Theoretically it could be possible to provide access to those matrices. But this would give you nothing in order to achieve the task you want. The quadrilateral transformation uses 2D homogenous matrices internally (3x3 matrix) and it will not give information about rotation around all 3 axis (you would need 4x4 homogenous matrix).

In order to estimate 3 rotations, you will need to study such algorithm like POSIT (search for "Model-based object pose in 25 lines of code" paper). This algorithm is implemented in OpenCV. Same is used in ARToolKit. In AForge.NET/GRATF it is still in progress.
With best regards,
Andrew


Interested in supporting AForge.NET Framework?
User avatar
andrew.kirillov
Site Admin, AForge.NET Developer
 
Posts: 3453
Joined: Fri Jan 23, 2009 9:12 am
Location: UK

Re: Quadrilateral transformation distance estimation

Postby moshe » Sat Feb 05, 2011 3:18 pm

Hello Andrew and thanks for the reply.

your comment is very interesting.
According to what you say, I cannot get the 3 rotations, but what about translation?
Without getting into math, intuitively I think that if you know the "real world" size of the glyph, and from the transformation you get the scale factor, you can extract the distance from the camera to the glyph.
Am I missing something here?

and another thing - if you do not have the full data in 6 degrees of freedom, does this mean you cannot draw a 3d object using this image?

Best regards,
Moshe.
moshe
 
Posts: 2
Joined: Sat Feb 05, 2011 9:42 am

Re: Quadrilateral transformation distance estimation

Postby andrew.kirillov » Sat Feb 05, 2011 4:14 pm

moshe wrote:Without getting into math, intuitively I think that if you know the "real world" size of the glyph, and from the transformation you get the scale factor, you can extract the distance from the camera to the glyph.

The GRATF provides quadrilateral corners for each detected glyph. So you may build the matrix on your own if you need.

moshe wrote:and another thing - if you do not have the full data in 6 degrees of freedom, does this mean you cannot draw a 3d object using this image?

As I said in my previous post, you need to use algorithms like POSIT to estimate 3D pose of the object. When you complete this step, you can draw 3D over it.
With best regards,
Andrew


Interested in supporting AForge.NET Framework?
User avatar
andrew.kirillov
Site Admin, AForge.NET Developer
 
Posts: 3453
Joined: Fri Jan 23, 2009 9:12 am
Location: UK




Return to GRATF