Implementation of the anisotropic iterative closest point (A-ICP) algoritm. More...

#include <mitkAnisotropicIterativeClosestPointRegistration.h>

Inheritance diagram for mitk::AnisotropicIterativeClosestPointRegistration:

Collaboration diagram for mitk::AnisotropicIterativeClosestPointRegistration:

Public Member Functions
	mitkClassMacroItkParent (AnisotropicIterativeClosestPointRegistration, itk::Object) static Pointer New()

Pointer	Clone () const

virtual void	SetMaxIterations (unsigned int _arg)

virtual void	SetThreshold (double _arg)

virtual void	SetFRENormalizationFactor (double _arg)

virtual void	SetSearchRadius (double _arg)

virtual void	SetMaxIterationsInWeightedPointTransform (double _arg)

virtual double	GetFRE ()

virtual unsigned int	GetNumberOfIterations ()

virtual void	SetTrimmFactor (double _arg)

virtual void	SetMovingSurface (itk::SmartPointer< Surface > _arg)

virtual void	SetFixedSurface (itk::SmartPointer< Surface > _arg)

virtual const Translation &	GetTranslation ()

virtual const Rotation &	GetRotation ()

void	SetCovarianceMatricesMovingSurface (CovarianceMatrixList &list)

void	SetCovarianceMatricesFixedSurface (CovarianceMatrixList &list)

void	Update ()

Protected Types
typedef itk::Matrix< double, 3, 3 >	CovarianceMatrix

typedef std::vector< CovarianceMatrix >	CovarianceMatrixList

typedef mitk::Vector3D	Translation

typedef CovarianceMatrix	Rotation

typedef std::pair< unsigned int, double >	Correspondence

typedef std::vector< Correspondence >	CorrespondenceList

Protected Member Functions
	AnisotropicIterativeClosestPointRegistration ()

	~AnisotropicIterativeClosestPointRegistration ()

void	ComputeCorrespondences (vtkPoints X, vtkPoints Z, vtkKdTreePointLocator *Y, const CovarianceMatrixList &sigma_X, const CovarianceMatrixList &sigma_Y, CovarianceMatrixList &sigma_Z, CorrespondenceList &correspondences, const double radius)

Protected Attributes
unsigned int	m_MaxIterations

double	m_Threshold

double	m_FRENormalizationFactor

double	m_SearchRadius

double	m_MaxIterationsInWeightedPointTransform

double	m_FRE

double	m_TrimmFactor

unsigned int	m_NumberOfIterations

itk::SmartPointer< Surface >	m_MovingSurface

itk::SmartPointer< Surface >	m_FixedSurface

itk::SmartPointer< WeightedPointTransform >	m_WeightedPointTransform

CovarianceMatrixList	m_CovarianceMatricesMovingSurface

CovarianceMatrixList	m_CovarianceMatricesFixedSurface

Translation	m_Translation

Rotation	m_Rotation

Detailed Description

Implementation of the anisotropic iterative closest point (A-ICP) algoritm.

This class implements the anisotropic interative closest point (A-ICP) algorithm presented in L. Maier-Hein et al. in "Convergent Iterative Closest-Point Algorithm to Accomodate Anisotropic and Inhomogenous Localization Error.", IEEE T Pattern Anal 34 (8), 1520-1532, 2012. The algorithm computes the optimal transformation to align two surfaces. In addition to the surfaces a list of covariance matrices is used as input for every surface. Each covariance matrix represents the error of a specific vertex in the Surface. The covariance matrices for each surface can be defined by the user, or calculated by the CovarianceMatrixCalculator. In addition a trimmed algorithm version is provided to compute the registration of partial overlapping surfaces. The algorithm needs a clean surface non manifold edges and without duplicated vertices. In addition vtkCleanPolyData can be used to ensure a correct Surface representation.

Note: The correspondence search is accelerated when OpenMP is enabled.

Example:

typedef itk::Matrix < double, 3, 3 > Matrix3x3;
typedef itk::Vector < double, 3 > Vector3;
typedef std::vector < Matrix3x3 > CovarianceMatrixList;
// compute the covariance matrices
mitk::CovarianceMatrixCalculator::Pointer matrixCalculator =
                                   mitk::CovarianceMatrixCalculator::New();
// compute the covariance matrices for the moving surface (X)
matrixCalculator->SetInputSurface(movingSurface);
matrixCalculator->ComputeCovarianceMatrices();
CovarianceMatrixList sigmas_X = matrixCalculator->GetCovarianceMatrices();
double meanVarX = matrixCalculator->GetMeanVariance();
// compute the covariance matrices for the fixed surface (Y)
matrixCalculator->SetInputSurface(fixedSurface);
matrixCalculator->ComputeCovarianceMatrices();
CovarianceMatrixList sigmas_Y = matrixCalculator->GetCovarianceMatrices();
double meanVarY = matrixCalculator->GetMeanVariance();
// the FRE normalization factor
double normalizationFactor = sqrt( meanVarX + meanVarY);
// A-ICP algorithm
mitk::AnisotropicIterativeClosestPointRegistration::Pointer aICP =
   mitk::AnisotropicIterativeClosestPointRegistration::New();
// set up parameters
aICP->SetMovingSurface(movingSurface);
aICP->SetFixedSurface(fixedSurface);
aICP->SetCovarianceMatricesMovingSurface(sigmas_X);
aICP->SetCovarianceMatricesFixedSurface(sigmas_Y);
aICP->SetFRENormalizationFactor(normalizationFactor);
// Trimming is enabled if a fator > 0.0 is set.
// 40 percent of the moving point set
// will be used for registration in this example.
// To disable trimming set the trim factor back to 0.0
aICP->SetTrimmFactor(0.4);
// run the algorithm
aICP->Update();
// retrieve the computed transformation
Matrix3x3 rotation = aICP->GetRotation();
Vector3 translation = aICP->GetTranslation();

Definition at line 118 of file mitkAnisotropicIterativeClosestPointRegistration.h.

Member Typedef Documentation

typedef std::pair<unsigned int, double> mitk::AnisotropicIterativeClosestPointRegistration::Correspondence

protected

Definition of a correspondeces, index and distance.

Definition at line 130 of file mitkAnisotropicIterativeClosestPointRegistration.h.

typedef std::vector<Correspondence> mitk::AnisotropicIterativeClosestPointRegistration::CorrespondenceList

protected

Definition of a list of correspondences.

Definition at line 132 of file mitkAnisotropicIterativeClosestPointRegistration.h.

typedef itk::Matrix<double, 3, 3> mitk::AnisotropicIterativeClosestPointRegistration::CovarianceMatrix

protected

Definition of a 3x3 covariance matrix.

Definition at line 122 of file mitkAnisotropicIterativeClosestPointRegistration.h.

typedef std::vector<CovarianceMatrix> mitk::AnisotropicIterativeClosestPointRegistration::CovarianceMatrixList

protected

Definition of a list of covariance matrices.

Definition at line 124 of file mitkAnisotropicIterativeClosestPointRegistration.h.

typedef CovarianceMatrix mitk::AnisotropicIterativeClosestPointRegistration::Rotation

protected

Definition of a 3x3 rotation matrix.

Definition at line 128 of file mitkAnisotropicIterativeClosestPointRegistration.h.

typedef mitk::Vector3D mitk::AnisotropicIterativeClosestPointRegistration::Translation

protected

Definition of a translation vector.

Definition at line 126 of file mitkAnisotropicIterativeClosestPointRegistration.h.

Constructor & Destructor Documentation

mitk::AnisotropicIterativeClosestPointRegistration::AnisotropicIterativeClosestPointRegistration ( )

protected

Definition at line 42 of file mitkAnisotropicIterativeClosestPointRegistration.cpp.

mitk::AnisotropicIterativeClosestPointRegistration::~AnisotropicIterativeClosestPointRegistration ( )

protected

Definition at line 57 of file mitkAnisotropicIterativeClosestPointRegistration.cpp.

Member Function Documentation

Pointer mitk::AnisotropicIterativeClosestPointRegistration::Clone ( ) const

void mitk::AnisotropicIterativeClosestPointRegistration::ComputeCorrespondences	(	vtkPoints *	X,
		vtkPoints *	Z,
		vtkKdTreePointLocator *	Y,
		const CovarianceMatrixList &	sigma_X,
		const CovarianceMatrixList &	sigma_Y,
		CovarianceMatrixList &	sigma_Z,
		CorrespondenceList &	correspondences,
		const double	radius
	)

protected

Method that computes the correspondences between the moving point set X and the fixed point set Y. The distances between the points are weighted with weight matrices that are computed from the covariances along the surfaces axes. This method implements the runtime optimization presented by L. Maier-Hein et al.. The correspondences are computed with the help of a kd tree. The correspondences are searched in a given radius in the euklidian space. Every correspondence found in this radius is weighted based on the covariance matrices and the best weighting will be used as a correspondence.

Parameters

X	The moving point set.
Z	The returned correspondences from the fixed point set.
Y	The fixed point set saved in a kd tree.
sigma_X	Covariance matrices belonging to the moving point set.
sigma_Y	Covariance matrices belonging to the fixed point set.
sigma_Z	Covariance matrices belonging to the correspondences found.
correspondences	Saved correspondences, in a pair containing the their index in Y and distance.
radius	The search radius used in in kd tree.

Definition at line 61 of file mitkAnisotropicIterativeClosestPointRegistration.cpp.

References mitk::AnisotropicRegistrationCommon::CalculateWeightMatrix(), max(), and mitk::New().

virtual double mitk::AnisotropicIterativeClosestPointRegistration::GetFRE ( )

virtual

Get the fiducial registration error (FRE).

virtual unsigned int mitk::AnisotropicIterativeClosestPointRegistration::GetNumberOfIterations ( )

virtual

Get the number of iterations used by the algorithm.

virtual const Rotation& mitk::AnisotropicIterativeClosestPointRegistration::GetRotation ( )

virtual

Returns the 3x3 rotation matrix computed by the algorithm.

virtual const Translation& mitk::AnisotropicIterativeClosestPointRegistration::GetTranslation ( )

virtual

Returns the 3x1 translation vector computed by the algorithm.

mitk::AnisotropicIterativeClosestPointRegistration::mitkClassMacroItkParent	(	AnisotropicIterativeClosestPointRegistration	,
		itk::Object
	)

void mitk::AnisotropicIterativeClosestPointRegistration::SetCovarianceMatricesFixedSurface ( CovarianceMatrixList & list )

inline

Set the covariance matrices of the fixed surface. The algorithm need the same amount of covariance and points available in the surface. The covariance matrix for every vertex in a Surface can be calculated by the CovarianceMatrixCalculator. It is also possible to define arbitrary matrices by hand.

Definition at line 298 of file mitkAnisotropicIterativeClosestPointRegistration.h.

void mitk::AnisotropicIterativeClosestPointRegistration::SetCovarianceMatricesMovingSurface ( CovarianceMatrixList & list )

inline

Set the covariance matrices of the moving surface. The algorithm need the same amount of covariance and points available in the surface. The covariance matrix for every vertex in a Surface can be calculated by the CovarianceMatrixCalculator. It is also possible to define arbitrary matrices by hand.

Definition at line 286 of file mitkAnisotropicIterativeClosestPointRegistration.h.

virtual void mitk::AnisotropicIterativeClosestPointRegistration::SetFixedSurface ( itk::SmartPointer< Surface > _arg )

virtual

Set fixed surface that includes the point set (Y).

virtual void mitk::AnisotropicIterativeClosestPointRegistration::SetFRENormalizationFactor ( double _arg )

virtual

Set the normalization factor for the fiducial registration error (FRE). The normalization factor is computed with the help of the mean variance of both CovarianceMatrixList that can be obtained when the covariance matrices are calculated with the CovarianceMatrixCalculator:

double FRENormalizationFactor = sqrt ( MeanVarianceX + MeanVarianceY );

if no FRE normalization is used the normalization factor is set to 1.0 as default value.

virtual void mitk::AnisotropicIterativeClosestPointRegistration::SetMaxIterations ( unsigned int _arg )

virtual

Set the maximum amount of iterations used by the algorithm.

virtual void mitk::AnisotropicIterativeClosestPointRegistration::SetMaxIterationsInWeightedPointTransform ( double _arg )

virtual

Set the maximim number of iterations used by the point based registration algorithm.

virtual void mitk::AnisotropicIterativeClosestPointRegistration::SetMovingSurface ( itk::SmartPointer< Surface > _arg )

virtual

Set moving surface that includes the point set (X).

virtual void mitk::AnisotropicIterativeClosestPointRegistration::SetSearchRadius ( double _arg )

virtual

Set search radius for the correspondence search.

virtual void mitk::AnisotropicIterativeClosestPointRegistration::SetThreshold ( double _arg )

virtual

Set the threshold used to terminate the algorithm.

virtual void mitk::AnisotropicIterativeClosestPointRegistration::SetTrimmFactor ( double _arg )

virtual

Factor that trimms the point set in percent for partial overlapping surfaces. E.g. 0.4 will use 40 precent of the point set. To enable the trimmed version a factor > 0 and < 1 must be set. The default value is 0.0.

void mitk::AnisotropicIterativeClosestPointRegistration::Update ( )

This method executes the algorithm.

Warning: The algorithm is only a simple calculation filter and can not be used in a mitk filter pipline.

Exceptions

Exception if the search radius was doubled more than 20 times to prevent endless loops. Re-run the with a different search radius that will find the correspondences.

Definition at line 135 of file mitkAnisotropicIterativeClosestPointRegistration.cpp.

References mitk::ProgressBar::AddStepsToDo(), AICPComp, mitk::ProgressBar::GetInstance(), max(), MITK_DEBUG, mitkThrow, mitk::New(), mitk::ProgressBar::Progress(), mitk::AnisotropicRegistrationCommon::PropagateMatrices(), and mitk::AnisotropicRegistrationCommon::TransformPoints().