Added a SVD based quadric optimisation for QE simplification

It allows to find the optimal position closest to a given point when
the quadrics are degenerated.
This commit is contained in:
Paolo Cignoni 2017-02-21 16:41:45 +01:00
parent 856f360af0
commit 43b22e4f42
1 changed files with 23 additions and 126 deletions

View File

@ -197,134 +197,31 @@ public:
} }
template <class ReturnScalarType>
bool MinimumClosestToPoint(Point3<ReturnScalarType> &x, const Point3<ReturnScalarType> &pt)
{
const double qeps = 1e-3;
Eigen::Matrix3d A;
Eigen::Vector3d be;
A << a[0], a[1], a[2],
a[1], a[3], a[4],
a[2], a[4], a[5];
be << -b[0]/2, -b[1]/2, -b[2]/2;
Eigen::JacobiSVD<Eigen::MatrixXd> svd(A, Eigen::ComputeThinU | Eigen::ComputeThinV);
Eigen::Vector3d s = svd.singularValues();
for(int i=1;i<3;++i)
if(s[i]/s[0] > qeps) s[i]=1/s[i];
else s[i]=0;
s[0]=1/s[0];
Eigen::Vector3d xp;
pt.ToEigenVector(xp);
Eigen::Vector3d xe = xp + (svd.matrixV()*s.asDiagonal()*(svd.matrixU().transpose())) *(be - A*xp);
// spostare..risolve un sistema 3x3 x.FromEigenVector(xe);
template<class FLTYPE>
bool Gauss33( FLTYPE x[], FLTYPE C[3][3+1] )
{
const FLTYPE keps = (FLTYPE)1e-3;
int i,j,k;
FLTYPE eps; // Determina valore cond.
eps = math::Abs(C[0][0]);
for(i=1;i<3;++i)
{
FLTYPE t = math::Abs(C[i][i]);
if( eps<t ) eps = t;
}
eps *= keps;
for (i = 0; i < 3 - 1; ++i) // Ciclo di riduzione
{
int ma = i; // Ricerca massimo pivot
FLTYPE vma = math::Abs( C[i][i] );
for (k = i + 1; k < 3; k++)
{
FLTYPE t = math::Abs( C[k][i] );
if (t > vma)
{
vma = t;
ma = k;
}
}
if (vma<eps)
return false; // Matrice singolare
if(i!=ma) // Swap del massimo pivot
for(k=0;k<=3;k++)
{
FLTYPE t = C[i][k];
C[i][k] = C[ma][k];
C[ma][k] = t;
}
for (k = i + 1; k < 3; k++) // Riduzione
{
FLTYPE s;
s = C[k][i] / C[i][i];
for (j = i+1; j <= 3; j++)
C[k][j] -= C[i][j] * s;
C[k][i] = 0.0;
}
}
// Controllo finale singolarita'
if( math::Abs(C[3-1][3- 1])<eps)
return false;
for (i=3-1; i>=0; i--) // Sostituzione
{
FLTYPE t;
for (t = 0.0, j = i + 1; j < 3; j++)
t += C[i][j] * x[j];
x[i] = (C[i][3] - t) / C[i][i];
}
return true; return true;
} }
template <class ReturnScalarType>
bool MinimumOld(Point3<ReturnScalarType> &x)
{
ReturnScalarType C[3][4];
C[0][0]=a[0]; C[0][1]=a[1]; C[0][2]=a[2];
C[1][0]=a[1]; C[1][1]=a[3]; C[1][2]=a[4];
C[2][0]=a[2]; C[2][1]=a[4]; C[2][2]=a[5];
C[0][3]=-b[0]/2;
C[1][3]=-b[1]/2;
C[2][3]=-b[2]/2;
return Gauss33(&(x[0]),C);
}
// determina il punto di errore minimo vincolato nel segmento (a,b)
bool Minimum(Point3<ScalarType> &x,Point3<ScalarType> &pa,Point3<ScalarType> &pb){
ScalarType t1,t2, t4, t5, t8, t9,
t11,t12,t14,t15,t17,t18,t25,t26,t30,t34,t35,
t41,t42,t44,t45,t50,t52,t54,
t56,t21,t23,t37,t64,lambda;
t1 = a[4]*pb.z();
t2 = t1*pa.y();
t4 = a[1]*pb.y();
t5 = t4*pa.x();
t8 = a[1]*pa.y();
t9 = t8*pa.x();
t11 = a[4]*pa.z();
t12 = t11*pa.y();
t14 = pa.z()*pa.z();
t15 = a[5]*t14;
t17 = a[2]*pa.z();
t18 = t17*pa.x();
t21 = 2.0*t11*pb.y();
t23 = a[5]*pb.z()*pa.z();
t25 = a[2]*pb.z();
t26 = t25*pa.x();
t30 = a[0]*pb.x()*pa.x();
t34 = 2.0*a[3]*pb.y()*pa.y();
t35 = t17*pb.x();
t37 = t8*pb.x();
t41 = pa.x()*pa.x();
t42 = a[0]*t41;
t44 = pa.y()*pa.y();
t45 = a[3]*t44;
t50 = 2.0*t30+t34+2.0*t35+2.0*t37-(-b[2]/2)*pa.z()-(-b[0]/2)*pa.x()-2.0*t42-2.0*t45+(-b[1]/2)*pb.y()
+(-b[0]/2)*pb.x()-(-b[1]/2)*pa.y();
t52 = pb.y()*pb.y();
t54 = pb.z()*pb.z();
t56 = pb.x()*pb.x();
t64 = t5+t37-t9+t30-t18+t35+t26-t25*pb.x()+t2-t1*pb.y()+t23;
lambda = (2.0*t2+2.0*t5+(-b[2]/2)*pb.z()-4.0*t9-4.0*t12-2.0*t15-4.0*t18+t21+2.0*t23+
2.0*t26+t50)/(-t45-a[3]*t52-a[5]*t54-a[0]*t56-t15-t42+t34-2.0*t12+t21-2.0*t4*pb.x()+
2.0*t64)/2.0;
if(lambda<0) lambda=0; else if(lambda>1) lambda = 1;
x = pa*(1.0-lambda)+pb*lambda;
return true;
}
}; };