[vtkusers] Point selection in Iterative closest point

Seth Gilchrist seth at mech.ubc.ca
Thu Jun 13 14:12:51 EDT 2013


Hello all,
I'm using ICP to register two surfaces together. In general I feel like I'm
getting poor results, given that my surfaces are already fairly well
aligned.

I've taken a look at the VTK ICP code and it seems that the problem might
be that it randomly samples points on the two surfaces and then aligns
based on finding the closest point on the target surface to its current
source point. The problem with this is that if the two surfaces only
partially overlap then the algorithm can select an edge point on the target
surface and force the surfaces to come together, even when there is good
match in the overlapping region.  To see what I mean, checkout figure 13
(page 6) in the document at
http://docs.happycoders.org/orgadoc/graphics/imaging/fasticp_paper.pdf

My question is: how can I modify the current ICP to select target points
that lie close to the source points along surface normals?  Also, is there
a wish list where I could add a request for the functionality to turn on or
off this kind of matching algorithm?

Thanks for your time,
Seth
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.vtk.org/pipermail/vtkusers/attachments/20130613/3383a666/attachment.htm>


More information about the vtkusers mailing list