This is the mother class of Optical Flow strategies: It implements different steps required for the iterative and parallel resolution of the linear system K.U = F. Notice that the parallelization is done on elements for OFFEM and on pixels for OFI. The function "stiffness" implements the calcul of K, i.e. the global stiffness tensor. It is calculated only one time and since it is based on the gradient of the img[0]. The function "forces" implements the calcul of F, i.e. the global forces vector. It is updated at each loop from img[1]* = img[1](x+U,y+V,z+W). Functions Fassembly and Kassembly are specific to OFFEM or OFI, thus are virtual pure, and implements within deeper class how K anf F assembly are done. The function "gradient" implements the point to point gradient. This function allows specifying how the gradient should be done in different case, i.e. centered, left or right one. The function "actu" implements the field actualization from new U vector as well as the difference between img[0] and img[1]* required for the further calcul of F. The function "regularize" implements a median filter used to enforce the smoothing of the solution vector "U" at each loop. It ensure a better convergency of the iterative calcul and is a first step in the implementation of a penality strategy (to do). The function "solve_cimg" is specific to OFFEM and OFI. Specifically when one use mask. It also allow using different solvers. More...

#include <Ccorrelation_opticalFlow.h>

Inheritance diagram for Ccorrelation_opticalFlow< T, Timg >:

Public Member Functions
virtual void	init (CParameterNetCDF &fp)

virtual void	exec (const Cimage< Timg > &oImage, const Cmesh< T, Timg > &oMesh, Cfield< T, Timg > &oField)

virtual void	assignG (const Cmesh< T, Timg > &oMesh, CImgList< T > &K, CImgList< T > &Ke)

virtual void	assignL (const Cmesh< T, Timg > &oMesh, CImgList< T > &Ve)=0

virtual void	reset (const int &thread, CImgList< T > &Ke)

virtual void	evalGN (const std::vector< T > &grad, const CImgList< T > &N, CImgList< T > &Ve)=0

virtual void	e_ref (const Cimage< Timg > &oImage, const Cmesh< T, Timg > &oMesh, const int &elem, std::vector< T > &ref)=0

virtual void	evalU (const CImgList< T > &solution, const Cmesh< T, Timg > &oMesh, const int &elem, const CImgList< T > &N, std::vector< T > &U)=0

virtual void	evalShape (const CImg< Timg > &list, const Cimage< Timg > &oImage, const Cfield< T, Timg > &oField, const std::vector< T > &ref, const std::vector< T > &lCoor, const std::vector< T > &gCoor, CImgList< T > &N)=0

virtual void	gradient (const Cimage< Timg > &oImage, const std::vector< T > &x, std::vector< T > &grad)

void	evalRes_loop (const Cimage< Timg > &oImage, const std::vector< T > &U, const std::vector< T > &x, T &residue, T &deformed)

void	evalRes_final (const Cimage< Timg > &oImage, const std::vector< T > &U, const std::vector< T > &x, T &residue, T &deformed)

virtual void	tensors (const int &thread, const CImgList< T > &Ve, const T &residue, CImgList< T > &Ke)

virtual void	assembly (const int &thread, const Cmesh< T, Timg > &oMesh, const int &elem, const CImgList< T > &Ke, CImgList< T > &K)=0

virtual void	metric (const Cimage< Timg > &oImage, const Cmesh< T, Timg > &oMesh, CImgList< T > &solution, Cfield< T, Timg > &oField, CImgList< T > &N, CImgList< T > &Ve, CImgList< T > &Ke, std::vector< T > &ref, std::vector< T > &offset, std::vector< T > &lCoor, std::vector< T > &gCoor, std::vector< T > &U, std::vector< T > &grad, T &residue, T &deformed, CImg< Timg > &list, CImg< int > &p, CImg< T > &current_res, CImgList< T > &K)

virtual void	solve (CImgList< T > &K, const Cmesh< T, Timg > &oMesh, CImgList< T > &solution)

virtual void	test (const CImgList< T > &solution, const double &epsilon, Cfield< T, Timg > &oField, T &previous_res, int &iter, int &converged)

virtual void	export_res (const Cimage< Timg > &oImage, const Cmesh< T, Timg > &oMesh, Cfield< T, Timg > &oField, CImgList< T > &N, std::vector< T > &ref, std::vector< T > &offset, std::vector< T > &lCoor, std::vector< T > &gCoor, std::vector< T > &U, T &residue, T &deformed, CImg< Timg > &list)

virtual void	store (Cfield< T, Timg > &oField, const Cimage< Timg > &oImage, const std::vector< T > gCoor, const std::vector< T > &U, const T &deformed, const T &residue)

Public Member Functions inherited from Ccorrelation< T, Timg >
virtual void	regularize (const Cmesh< T, Timg > &oMesh, const Cimage< Timg > &oImage, CImgList< T > &solution)

void	count (const int &increment, const int &dim, int &counter)

Public Attributes
std::string	m_shape_type

std::string	m_interpLoop

std::string	m_interpFinal

double	m_epsilon

std::vector< T >	m_ref

Public Attributes inherited from Ccorrelation< T, Timg >
std::string	class_name

std::string	m_correl_type

std::string	m_correl_name

T	m_current_res

int	m_threadNB

int	_3D

int	m_medianFilter

bool	m_verbose

Detailed Description

template<typename T, typename Timg>
class Ccorrelation_opticalFlow< T, Timg >

This is the mother class of Optical Flow strategies: It implements different steps required for the iterative and parallel resolution of the linear system K.U = F. Notice that the parallelization is done on elements for OFFEM and on pixels for OFI. The function "stiffness" implements the calcul of K, i.e. the global stiffness tensor. It is calculated only one time and since it is based on the gradient of the img[0]. The function "forces" implements the calcul of F, i.e. the global forces vector. It is updated at each loop from img[1]* = img[1](x+U,y+V,z+W). Functions Fassembly and Kassembly are specific to OFFEM or OFI, thus are virtual pure, and implements within deeper class how K anf F assembly are done. The function "gradient" implements the point to point gradient. This function allows specifying how the gradient should be done in different case, i.e. centered, left or right one. The function "actu" implements the field actualization from new U vector as well as the difference between img[0] and img[1]* required for the further calcul of F. The function "regularize" implements a median filter used to enforce the smoothing of the solution vector "U" at each loop. It ensure a better convergency of the iterative calcul and is a first step in the implementation of a penality strategy (to do). The function "solve_cimg" is specific to OFFEM and OFI. Specifically when one use mask. It also allow using different solvers.

Parameters

[in]	m_shape_type	: <int>. This is the shape function type. For OFFEM, Q4 is implemented, and for OFI, rigid_body 2D and 3D, homogeneous 2D and 3D, dilatation 2D and 3D, brasilian 2D, blade 2D are implemented.
[in]	m_epsilon	: <double>. This is the limit rate of convergence. Under this value calcul is considered as converged within a minimum.

Member Function Documentation

template<typename T , typename Timg >

void Ccorrelation_opticalFlow< T, Timg >::evalRes_final	(	const Cimage< Timg > &	oImage,
		const std::vector< T > &	U,
		const std::vector< T > &	x,
		T &	residue,
		T &	deformed
	)

inline

the displacement U[dim] in physical unit is divided by the Ratio Pix/millimeter considering the current size of super pixel.//

template<typename T , typename Timg >

void Ccorrelation_opticalFlow< T, Timg >::evalRes_loop	(	const Cimage< Timg > &	oImage,
		const std::vector< T > &	U,
		const std::vector< T > &	x,
		T &	residue,
		T &	deformed
	)

inline

the displacement U[dim] in physical unit is divided by the Ratio Pix/millimeter considering the current size of super pixel.//

template<typename T , typename Timg >

virtual void Ccorrelation_opticalFlow< T, Timg >::gradient	(	const Cimage< Timg > &	oImage,
		const std::vector< T > &	x,
		std::vector< T > &	grad
	)

inlinevirtual

implements border conditions (global centered gradient, but right and left one for left and right borders respectively). An other check is done for pixels, in an element, but at the border of a mask.

Reimplemented in Ccorrelation_opticalFlow_integrated< T, Timg >.

template<typename T , typename Timg >

virtual void Ccorrelation_opticalFlow< T, Timg >::init ( CParameterNetCDF & fp )

inlinevirtual

for an optical-flow correlation, required attributes are the shape function, the convergence limit speed

Implements Ccorrelation< T, Timg >.

Reimplemented in Ccorrelation_opticalFlow_fem_gradient< T, Timg >, Ccorrelation_opticalFlow_integrated< T, Timg >, Ccorrelation_opticalFlow_fem< T, Timg >, Ccorrelation_opticalFlow_integrated_block< T, Timg >, Ccorrelation_opticalFlow_integrated_block_A< T, Timg >, and Ccorrelation_opticalFlow_integrated_block_A< T, Timg >.

The documentation for this class was generated from the following file:

/home/rian/Documents/PostDoc/LML/Yadics/YaDICs_Project/YaDICs_V04.14.a/Ccorrelation_opticalFlow.h

Public Member Functions

Public Attributes

Detailed Description

template<typename T, typename Timg> class Ccorrelation_opticalFlow< T, Timg >

Member Function Documentation

template<typename T, typename Timg>
class Ccorrelation_opticalFlow< T, Timg >