void fuseRange(BuildData data, int first, int last)
{
int colCount = last - first;
var sum = Vector3.Zero;
for (int i = 0; i < colCount; i++)
{
sum += data.points[data.indices[first + i]].point;
}
var mean = sum / colCount;
var NN = new MathNet.Numerics.LinearAlgebra.Single.DenseMatrix(3, colCount);
var min = new Vector3(float.MaxValue);
var max = new Vector3(float.MinValue);
for (int i = 0; i < colCount; i++)
{
var v = data.points[data.indices[first + i]].point;
var p = v - mean;
NN.At(0, i, p.X);
NN.At(1, i, p.Y);
NN.At(2, i, p.Z);
min = Vector3.Min(min, v);
max = Vector3.Max(max, v);
}
Vector3 box = max - min;
float boxDim = GetAt(box, MaxDim(box));
// drop box if it is too large
if (boxDim > maxBoxDim)
{
data.unfitPointsCount += colCount;
return;
}
// compute covariance
var C = NN.TransposeAndMultiply(NN);
var eigen = C.Evd();
// Ensure that the matrix is suited for eigenvalues calculation
if (eigen.Rank < 2)
{
data.unfitPointsCount += colCount;
return;
}
var normal = computeNormal(eigen.EigenValues(), eigen.EigenVectors());
/*T densitie = 0;
* if(keepDensities)
* densitie = SurfaceNormalDataPointsFilter::computeDensity(NN);*/
// Filter points randomly
if (samplingMethod == SamplingMethod.RandomSampling)
{
for (int i = 0; i < colCount; i++)
{
float x = (float)r.NextDouble();
if (x < ratio)
{
// Keep points with their descriptors
int k = data.indices[first + i];
// Mark the indices which will be part of the final data
data.pointsToKeep.Add(new DataPoint
{
point = data.points[k].point,
normal = normal,
});;
}
}
}
else if (samplingMethod == SamplingMethod.Bin)
{
// Use the average and norm
data.pointsToKeep.Add(new DataPoint
{
point = mean,
normal = normal
});
}
}
