Solving for optimal alignment of 3d polygonal mesh - optimization

I'm trying to implement a geometry templating engine. One of the parts is taking a prototypical polygonal mesh and aligning an instantiation with some points in the larger object.
So, the problem is this: given 3d point positions for some (perhaps all) of the verts in a polygonal mesh, find a scaled rotation that minimizes the difference between the transformed verts and the given point positions. I also have a centerpoint that can remain fixed, if that helps. The correspondence between the verts and the 3d locations is fixed.
I'm thinking this could be done by solving for the coefficients of a transformation matrix, but I'm a little unsure how to build the system to solve.
An example of this is a cube. The prototype would be the unit cube, centered at the origin, with vert indices:
4----5
|\ \
| 6----7
| | |
0 | 1 |
\| |
2----3
An example of the vert locations to fit:
v0: 1.243,2.163,-3.426
v1: 4.190,-0.408,-0.485
v2: -1.974,-1.525,-3.426
v3: 0.974,-4.096,-0.485
v5: 1.974,1.525,3.426
v7: -1.243,-2.163,3.426
So, given that prototype and those points, how do I find the single scale factor, and the rotation about x, y, and z that will minimize the distance between the verts and those positions? It would be best for the method to be generalizable to an arbitrary mesh, not just a cube.

Assuming you have all points and their correspondences, you can fine-tune your match by solving the least squares problem:
minimize Norm(T*V-M)
where T is the transformation matrix you are looking for, V are the vertices to fit, and M are the vertices of the prototype. Norm refers to the Frobenius norm. M and V are 3xN matrices where each column is a 3-vector of a vertex of the prototype and corresponding vertex in the fitting vertex set. T is a 3x3 transformation matrix. Then the transformation matrix that minimizes the mean squared error is inverse(V*transpose(V))*V*transpose(M). The resulting matrix will in general not be orthogonal (you wanted one which has no shear), so you can solve a matrix Procrustes problem to find the nearest orthogonal matrix with the SVD.
Now, if you don't know which given points will correspond to which prototype points, the problem you want to solve is called surface registration. This is an active field of research. See for example this paper, which also covers rigid registration, which is what you're after.

If you want to create a mesh on an arbitrary 3D geometry, this is not the way it's typically done.
You should look at octree mesh generation techniques. You'll have better success if you work with a true 3D primitive, which means tetrahedra instead of cubes.
If your geometry is a 3D body, all you'll have is a surface description to start with. Determining "optimal" interior points isn't meaningful, because you don't have any. You'll want them to be arranged in such a way that the tetrahedra inside aren't too distorted, but that's the best you'll be able to do.

Related

CGAL Mean_curvature_flow_skeletonization contract_until_convergence function produces branches that does not exist in input polygon

I use contract_until_convergence function from CGAL Mean_curvature_flow_skeletonization to produce skeleton from input polygon.
https://doc.cgal.org/latest/Surface_mesh_skeletonization/classCGAL_1_1Mean__curvature__flow__skeletonization.html
In some cases the skeleton creates branches (see top of the image above, skeleton in red color) that does not exist in input polygon. Is there some parameters to set to prevent this ?
using Skeletonization = CGAL::Mean_curvature_flow_skeletonization<Polyhedron>;
Skeletonization mean_curve_skeletonizer(polyhedron);
mean_curve_skeletonizer.contract_until_convergence();
There are two parameters controlling the quality of the skeleton:
quality_speed_tradeoff()
medially_centered_speed_tradeoff()
Also one thing that affect the skeleton is the sampling of the input surface that is used to compute Voronoi poles. In the original papers, it is said: Given a sufficiently good sampling, the Voronoi poles [ACK00] form a provably convergent sampling of the medial axis.
[ACK00]AMENTAN., CHOIS., KOLLURIR. K.: The powercrust, unions of balls, and the medial axis transform. Computational Geometry: Theory and Applications 19(2000), 127–153.3
You can use the function isotropic_remeshing with a sufficiently small target edge length to improve the Voronoi pole computation.

How to map the node identities of my resulting surface mesh generated from Poisson_surface_reconstruction_3 into my starting point sets?

thanks for reading this question. My title is basically what I'm trying to achieve. I did a poisson surface mesh generation using Poisson_surface_reconstruction_3(cgal). I can't figure out how to map the node identities of my resulting surface mesh into my starting point sets?
The output of my poisson surface generation is produced by the following lines:
CGAL::facets_in_complex_2_to_triangle_mesh(c2t3, output_mesh);
out << output_mesh;
In my output file, there are some x y z coordinates, followed by a set of 3 integers each line, I think they indicates which nodes form a delaunay triangle. The problem is that the output points do not correspond to my initial point set, since not any x y z value match to any of my original points. Yet I'm trying to figure out which points are forming a delaunay triangles in my original point set.
Could someone suggest me how can I do this in cgal?
Many thanks.
The poisson recontruction algorithm consist in meshing an implicit function that somehow fits you input points. In practice, it means that you input point will no belong to the set of points of the output surface, and won't even lie exactly on triangles of the output surface. However, they should not be too far from the output surface (except if you have some really sparse sampling parts).
What you can do to locate your input points with the output surface is to use the function closest_point_and_primitive() from the AABB-tree class.
Here is an example of how to build the tree from a mesh.

Computing Minkowski Difference For Circles and Convex Polygons

I'm needing to implement a Minkowski sum function that can return the Minkowski sum of either 2 circles, 2 convex polygons or a circle and a convex polygon. I found this thread that explained how to do this for convex polygons, but I'm not sure how to do this for a circle and polygon. Also, how would I even represent the answer?! I'd like the algorithm to run in O(n) time but beggars can't be choosers.
Circle is trivial -- just add the center points, and add the radii. Circle + ConvexPoly is nearly as simple: move each segment perpendicularly outward by the circle radius, and connect adjacent segments with circular arcs centered at the original poly vertices. Translate the whole by the circle center point.
As for how you represent the answer: Well, it depends on what you want to do with it. You could convert it to a NURBS if you just want to draw it with a vector drawing library. You could approximate the circular arcs with polylines if you just want a polygonal approximation. Or you might store it as is -- "this polygon, expanded by such-and-such a radius". That would be the best choice for things like raycasting, for instance. Or as a compromise, you could connect adjacent segments linearly instead of with circular arcs, and store it as the union of the (new) convex polygon and a list of circles at the vertices.
Oh, about ConvexPoly + ConvexPoly. That's the trickiest one, but still straightforward. The basic idea is that you take the list of segment vectors for each polygon (starting from some particular extremal point, like the point on each poly with the lowest X coordinate), then merge the two lists together, keeping it sorted by angle. Sum the two points you started with, then apply each vector from the merged vector list to produce the other points.

Reconstruct surface from 3D triangular meshes

I have a 3D model, which consists of the 3D triangular meshes. I want to partition the meshes into different groups. Each group represents a surface, such as a planar face, cylindrical surface. This is something like surface recognition/reconstruction.
The input is a set of 3D triangular meshes. The output is the mesh segmentations per surface.
Is there any library meets my requirement?
If you want to go into lots of mesh processing, then the point cloud library is a good idea, but I'd also suggest CGAL: http://www.cgal.org for more algorithms and loads of structures aimed at meshes.
Lastly, the problem you describe is most easily solved on your own:
enumerate all vertices
enumerate all polygons
create an array of ints with the size of the number of vertices in your "big" mesh, initialize with 0.
create an array of ints with the size of the number of polygons in your "big" mesh, initialize with 0.
initialize a counter to 0
for each polygon in your mesh, look at its vertices and the value that each has in the array.
if the values for each vertex are zero, increase counter and assign to each of the values in the vertex array and polygon array correspondingly.
if not, relabel all vertices and polygons with a higher number to the smallest, non-zero number.
The relabeling can be done quickly with a look up table.
This might save you lots of issues interfacing your code to some library you're not really interested in.
You should have a look at the PCL library, it has all these features and much more: http://pointclouds.org/

Least Squared constrained for Rototranslation

I want to detect the best rototraslation matrix between two set of points.
The second set of points is the same of the first, but rotated, traslated and affecteb by noise.
I tried to use least squared method by obviously the solution is usually similar to a rotation matrix, but with incompatible structure (for example, where i should get a value that represents the cosine of an angle i could get a value >1).
I've searched for the Constrained Least Squared method but it seems to me that the constrains of a rototraslation matrix cannot be expressed in this form.
In this PDF i've stated the problem more formally:
http://dl.dropbox.com/u/3185608/minquad_en.pdf
Thank you for the help.
The short answer: What you will need here is "Principal Component Analysis".
Apply this to both sets of points centered at their respective centers of mass. The PCA will effectively give you a rotation matrix for each aligned to the data set principal components. Multiplying the inverse matrix of the original set by the new rotation will give you a matrix that takes the old (centered) set to the new. Inverse translations and translations can similarly be applied to the rotation to create a homogeneous matrix that maps the one set to the other.
The book PRINCE, Simon JD. Computer vision: models, learning, and inference. Cambridge University Press, 2012.
gives, in Appendix "B.4 Reparameterization", some info about how to constrain a matrix to be a rotation matrix.
It seems to me that your problem has also a solution based on SVD: see the Kabsch algorithm also described by Olga Sorkine-Hornung and Michael Rabinovich in
Least-Squares Rigid Motion Using SVD and, more practically, by Nghia Kien Ho in FINDING OPTIMAL ROTATION AND TRANSLATION BETWEEN CORRESPONDING 3D POINTS.