[vtkusers] vtkSurfaceReconstructionFilter Alternatives

[Scott McMillan]

Hello All.

I am attempting to reconstruct a surface from an unstructured point 
cloud (which does describe a surface, though with noise), and am also 
very new to VTK. I'm hoping that someone might be able to suggest some 
improvements to the current methods I am using, because there are some 
issues.

The results I have obtained using vtkSurfaceReconstructionFilter have 
been very poor. I.e., the output doesn't look much like the model at all 
in a lot of circumstances. I don't think this is due to anything I am 
doing wrong either (though admittedly the documentation does mention 
that it needs nice input data). I modified the cactus input for the 
vtkSurfaceReconstructionFilter example in a text editor to have a single 
vertex offset from its correct position, and this produced a lot of 
noise all around the cactus. This makes me think that this filter is not 
ever going to be useful if there is noise in the data.

I have also tried some other approaches such as using Gaussian splatter 
to try and reproduce a surface, but the results have not been very 
satisfactory (a bit like you would expect).

vtkPowerCrustSurfaceReconstruction seems to offer the most promise (from 
http://www.sq3.org.uk/powercrust/ ) however on 200 000 points there's no 
hope for my Linux machine with 1 GB of RAM to avoid a lot of swapping 
due to the high memory usage (and hence very poor performance). I would 
also like to be able to process data with more points than this, but 
given the memory consumed for 200 000 points I have my doubts as to 
whether 500 000 or so points would even fit into the amount of memory 
available to a 32 bit process.

If someone could point out some alternatives to me, or even just tell me 
that there are no alternatives available in VTK at the moment, that 
would be very helpful to me.

Regards,
Scott McMillan.