[Insight-developers] kanade-lucas-tomasi tracking/registration

[Luis Ibanez]

Hi Richard,

 From the description of the Kanade-Lucas-Tomasi method in the page that
you posted, it seems that ITK has indeed all the basic components needed
for implementing this method.

The Taylor series expansion of an image to linear terms seems to be
equivalent to computing the gradient of the image. Which ITK does
as a default for most of the image metrics implemented.

The metric that is used in the KLT method is implemented in ITK by
the MeanSquareImageToImage metric, and the optimizer is equivalent
to the GradientDescent optimizer.

Notice that ITK is much more generic than the KLT method. In ITK the
derivative of the Metric with respect to the Transform parameters is
computed by applying the chain rule, and multiplying the image gradient
by the Jacobian of the Transform.

If your final purpose is to actually track objects, you may find better
options for doing this in ITK than the simple mapping of concepts from
the KLT method.  For example, in the interest of computational speed
you probably will be interested in trying the PointSet to Image regis-
tration since it makes possible to select significant pixels from one
image and track them in the other image.


You will find all these components described in detail in the ITK
Software Guide:

         http://www.itk.org/ItkSoftwareGuide.pdf

In the Image Registration Chapter.



Please let us know if you find any difficulty mapping the concepts
from the KLT method into ITK. We will be happy to give your more
details.


     Thanks


        Luis



------------------
Richard Beare wrote:
> Hi everyone,
> 
> I was wondering whether an equivalent of the kanade-lucas-tomasi
> tracking is available somewhere in the itk registration framework?
> 
> The derivation involves the Taylor series expansion of the image to
> linear terms, differentiating wrt the transform and solving. I suspect
> that this is probably equivalent to some of the fancier optimizers
> already in ITK, but can anyone suggest which ones?
> 
> Some documentation is available at:
> 
> http://www.ces.clemson.edu/~stb/klt/
> 
> All suggestions welcome.
> 
> Thanks
> _______________________________________________
> Insight-developers mailing list
> Insight-developers at itk.org
> http://www.itk.org/mailman/listinfo/insight-developers
> 
>