fine registration of 2 facial scans

[srubi]

Hello

I am working on my thesis related to 3D simulation of orthognatic surgery. The aim is to compare 2 3D models, which represent one object - face, but one of these models is slightly different fron the second one in some regions. The first one is a mesh (simulation of surgery) - reference model and the second one is a cloud - control model. Hypotheticaly, there is one region of these 2 objects (their surfaces), which is identical for both of them, it is forehead. I need to use this unchanged region to finely register these 2 models together (only over foreheads) and than measure distances between the rest regions.
And the question is, whether is the following method of registration suitable?

1. point-pair registration - selection of several,more or less coresponding points in the region of forehead, performing rough registration
2. fine registration using ICP algorithm

Does ICP algorithm use only regional points of forehead to finely register 2 models?

Thank You for your help
Martin

[daniel]

Hi,

the workflow is good but for the second part you'll have to either use a quite low 'overlap' percentage (roughly the number of points on the forehead relatively to the total number of points).

If this is not sufficient you can also manually segment the forehead (on the aligned cloud), use it for fine registration and then apply the same transformation to the whole cloud. But hopefully the 'overlap' parameter should be sufficient.

[srubi]

Thank You very much,
the first method seems to work sufficiently, regarding the second method - everytime I have tried to apply the transformational matrix of forehead to the whole facial scan I obtained dislocated registration.
Is it possible to segment registrated scans into several single patches and subsequently count the mean distances between the scans?
Thank You
Martin

[daniel]

There's no method to do this automatically right now. In command line mode you can crop the cloud and register the resulting subset of points (so you could create a batch process). But you would have to compute the mean value "manually". I guess it would be better to code a new function/plugin (if you or someone you know have some dev skills ).