[ITK Community] ITK

[Luis Ibanez]

Hi Francisco,

The ICP implementation in ITK uses the Levenberg Marquardt optimizer,
which in ITK is a wrapper of the VNL class vnl_levenberg_marquardt,
that in turn is a wrapper around FORTRAN routines.

You may find interesting to look at:
https://github.com/InsightSoftwareConsortium/ITK/blob/master/Modules/ThirdParty/VNL/src/vxl/core/vnl/algo/vnl_levenberg_marquardt.cxx#L458

that is the method:

vnl_matrix<double> const& vnl_levenberg_marquardt::get_JtJ()

that computes the inverse of the Covariance matrix.

One could get access to the vnl_levenberg_marquardt optimizer object
by adding a Get method to the itkLevenbergMarquardtOptimizer, and
returning the InternalOptimizerType m_VnlOptimier member variable:

https://github.com/InsightSoftwareConsortium/ITK/blob/master/Modules/Numerics/Optimizers/include/itkLevenbergMarquardtOptimizer.h#L89



  Hope this helps,

      Best

         Luis



On Wed, Nov 20, 2013 at 5:29 AM, Francisco Javier Pérez Grau <
fjperezgrau at gmail.com> wrote:

> Hello Luis,
> My name is Francisco Perez, and I am currently involved in a robotics
> project for navigation based on 3D imaging sensors such as Kinect. I am
> looking for ICP implementations in order to perform registration of
> consecutive point clouds, and found 'ITK' so I am playing a little bit with
> it, it looks promising! Congratulations on this job!
> I have one concern though, because my idea is to generate odometry and
> integrate with other sensors, and I don't know if I can obtain the
> covariance matrix of this calculation. How could I obtain this matrix? Is
> it calculated at all?
> Thank you in advance,
> Francisco.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/community/attachments/20131124/1785e476/attachment.html>