[vtkusers] problems with ICP alignement

[Tim Hutton]

Hi Jonathan,

In general there is no guarantee that ICP will converge to the correct 
alignment. Tricks that might help on some types of data include:

- start by aligning centroids (vtkICP has this be default)
- start by aligning principal axes
- start from multiple positions / orientations and use the result that 
gives the smallest mean distance

hth,

Tim.

At 17:48 07/08/2003 +0200, Jonathan.Bailleul wrote:
>"Jonathan.Bailleul" wrote:
> >
> > Dear all,
> >
> > I'm using ICP to align polydata of very similar shape but various
> > positions and orientation. In general, this works fine for the shape I
> > study, but here is a case where ICP fails!
> > (http://www.greyc.ismra.fr/~bailleul/These/ICPfail-start.jpg &
> > http://www.greyc.ismra.fr/~bailleul/These/ICPfail-end.jpg)
> >
> > Please note that in the general case, I might also be obliged to adjust
> > the scale factor of the mesh to align in order to converge to a relevant
> > result: in that one, no adjustment managed to make it.
> >
> > How could I do to have it work in all cases? Adjusting the initial
> > orientation of the shapes (how?)?
> >
> > In advance, thanks a lot for your comments
>
>
>I just found a trick to help me out of my own trap, but nothing to be
>proud of.
>The shape to align is always the same in all my tests, and the reference
>shape does not vary too much in position and orientation (since
>extracted from brains that are globally aligned). Even that little
>variation is enough to have the ICP unstable, but the most important
>difference is in the orientation of the shape to align regarding to the
>reference ones.
>So, for one case where I managed to align the shape, I replaced the mean
>shape by this one with good success in other cases!
>
>But since this is far from satisfying (manual operations required), I'd
>be glad to benefit from your suggestions.
>
>
>Best regards,
>
>
>--
>-----------------------------------
>Jonathan BAILLEUL, Doctorant
>GREYC Image - Université de Caen
>http://www.greyc.ismra.fr/~bailleul
>_______________________________________________
>This is the private VTK discussion list.
>Please keep messages on-topic. Check the FAQ at: 
><http://public.kitware.com/cgi-bin/vtkfaq>
>Follow this link to subscribe/unsubscribe:
>http://www.vtk.org/mailman/listinfo/vtkusers