Nonconvex 3d model algorithm

Question

I have cloud of 3d points (relief) and I need a simple algorithm of creating nonconvex 3d model. Could you help me?

Answer 1

A quick and rather easy to implement algorithm for getting a decent hull of triangles around the points (outside and inside!) is based on the "Marching Cubes" algorithm:

• define a grid (minimal x,y,z values + grid resolutions in x,y,z direction + number of points in x,y,z direction)
• initialize a value for each grid point to zero
• "rasterize" the given points (round the coordinates to the nearest grid point); for the trivial version just set the corresponding grid point's value to 1
• Now you can polygonize each basic cube of the grid with the "Marching Cubes" algorithm with an isolevel of 0.9.

These code snippets might help you. For initializing:

union XYZ {
    struct {
        double x,y,z;
    };
    double coord[3];
};
typedef std::vector<XYZ> TPoints;
TPoints points;
// get points from file or another data structure

// initialize these values for your needs (the bounding box of the points will help)    
XYZ coordsMin, coordsMax;
XYZ voxelSize; // grid resolution in each coordinate direction
int gridSize; // number of grid points, here the same for each coordinate direction

int gridSize2 = gridSize*gridSize;
char* gridVals = new char[gridSize2*gridSize];
for (int i=0; i<gridSize; ++i)
for (int j=0; j<gridSize; ++j)
for (int k=0; k<gridSize; ++k)
    gridVals[i*gridSize2+j*gridSize+k] = 0;

for (size_t i=0; i<points,size(); ++i)
{
    XYZ const& p = points[i];
    int gridCoords[3];
    for (int c=0; c<3; ++c)
        gridCoords[c] = (int)((p.coord[c]-coordsMin.coord[c])/voxelSize.coord[c]);
    int gridIdx = gridCoords[0]*gridSize2 + gridCoords[1]*gridSize + gridCoords[2];
    gridVals[gridIdx] = 1;
}

And then for computing the triangles based on the function Polygonize from the cited implementation:

const double isolevel = 0.9;

TRIANGLE triangles[5]; // maximally 5 triangles for each voxel
int cellCnt = 0;
for (int i=0; exportOk && i<gridSize-1; ++i)
for (int j=0; exportOk && j<gridSize-1; ++j)
for (int k=0; exportOk && k<gridSize-1; ++k)
{
    GRIDCELL cell;
    for (int cornerIdx=0; cornerIdx<8; ++cornerIdx)
    {
        XYZ& corner = cell.p[cornerIdx];
        // the function Polygonize expects this strange order of corner indexing ...
        // (see http://paulbourke.net/geometry/polygonise/)
        int xoff = ((cornerIdx+1)/2)%2;
        int yoff = (cornerIdx/2)%2;
        int zoff = cornerIdx/4;
        corner.x = (i+xoff)*voxelSize.x + coordsMin.x;
        corner.y = (j+yoff)*voxelSize.y + coordsMin.y;
        corner.z = (k+zoff)*voxelSize.z + coordsMin.z;
        cell.val[cornerIdx] = gridVals[(i+xoff)*gridSize2+(j+yoff)*gridSize+k+zoff];
    }
    int triangCnt = Polygonise(cell, isolevel, triangles);
    triangCntTotal += triangCnt;
    triangCntTotal += triangCnt;
    for (int t=0; t<triangCnt; ++t)
    {
        TTriangle const& triangle = triangles[t].p;
        ExportTriangle(triangle);
    }
}

This did the trick for me in an industrial application.