lti::transformEstimator Class Reference

A parent class for estimating a transform from sets of points. More...

#include <ltiTransformEstimator.h>

Inheritance diagram for lti::transformEstimator:

[legend]

Collaboration diagram for lti::transformEstimator:

[legend]

List of all members.

Classes
class	parameters
	The parameters for the class transformEstimator. More...
Public Member Functions
	transformEstimator ()
	transformEstimator (const parameters &par)
	transformEstimator (const transformEstimator &other)
virtual	~transformEstimator ()
virtual const char *	getTypeName () const
virtual bool	apply (const matrix< ipoint > &src, fvector &dest) const =0
virtual bool	apply (const matrix< ipoint > &src, fvector &dest, fvector &error) const =0
virtual bool	apply (const matrix< ipoint > &src, fvector &dest, const ivector &indices, int numCorrespondences) const =0
virtual bool	apply (const matrix< ipoint > &src, fvector &dest, fvector &error, const ivector &indices, int numCorrespondences) const =0
virtual bool	apply (const matrix< fpoint > &src, fvector &dest) const =0
virtual bool	apply (const matrix< fpoint > &src, fvector &dest, fvector &error) const =0
virtual bool	apply (const matrix< fpoint > &src, fvector &dest, const ivector &indices, int numCorrespondences) const =0
virtual bool	apply (const matrix< fpoint > &src, fvector &dest, fvector &error, const ivector &indices, int numCorrespondences) const =0
virtual bool	apply (const matrix< dpoint > &src, dvector &dest) const =0
virtual bool	apply (const matrix< dpoint > &src, dvector &dest, dvector &error) const =0
virtual bool	apply (const matrix< dpoint > &src, dvector &dest, const ivector &indices, int numCorrespondences) const =0
virtual bool	apply (const matrix< dpoint > &src, dvector &dest, dvector &error, const ivector &indices, int numCorrespondences) const =0
virtual bool	computeResidual (const matrix< fpoint > &src, const fvector &transform, fvector &dest) const =0
virtual bool	computeResidual (const matrix< dpoint > &src, const dvector &transform, dvector &dest) const =0
virtual int	minNumberCorrespondences () const =0
virtual int	minCorrespondenceDimension () const =0
virtual int	maxCorrespondenceDimension () const =0
virtual bool	denormalize (fvector &srcdest, const vector< fpoint > &scale, const vector< fpoint > &shift) const =0
virtual bool	denormalize (dvector &srcdest, const vector< dpoint > &scale, const vector< dpoint > &shift) const =0
transformEstimator &	copy (const transformEstimator &other)
transformEstimator &	operator= (const transformEstimator &other)
virtual functor *	clone () const =0
const parameters &	getParameters () const

Detailed Description

A parent class for estimating a transform from sets of points.

For automtic outlier detection please use one of the Monte Carlo estimators, which call a transform estimator in their main loop. The Monte Carlo estimators perform data normalization on request, whereas this class does not.

If you derive a class please remember to support the paramter for returning a squared error measure (e.g. squared residual) instead of the error (e.g. residual), and thus spare taking the square root.

See also:: lti::homography8DOFEstimator, lti::homography9DOFEstimator, lti::fMatrixEstimator, lti::monteCarloEstimator

Also planned:

estimating a homography together with a radial distortion

Constructor & Destructor Documentation

lti::transformEstimator::transformEstimator ( )

Default constructor.

lti::transformEstimator::transformEstimator ( const parameters & par )

Construct a functor using the given parameters.

lti::transformEstimator::transformEstimator ( const transformEstimator & other )

Copy constructor.

Parameters:

other

the object to be copied

virtual lti::transformEstimator::~transformEstimator ( ) [virtual]

Destructor.

Member Function Documentation

virtual bool lti::transformEstimator::apply	(	const matrix< dpoint > &	src,
		dvector &	dest,
		dvector &	error,
		const ivector &	indices,
		int	numCorrespondences
	)			const `[pure virtual]`

Estimates a transform from the supplied point sets, whereby only the points specified in the index vector are considered.

Usually this method calls the apply without the residual first, and then computes the residual.

All points of one point set give a matrix row, whereas all elements of a specifec correspondence stand in a matrix column.

Parameters:

	src	matrix<dpoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns.
	dest	dvector the estimated transform.
	error	dvector containing the deviation of each point from the estimated transform. Usually this is the residual or elementwise squared residual.
	indices	ivector with the indices of the relevant points.
	numCorrespondences	the first numCorrespondences indices are considered.

Returns:: true if the transform could be estimated and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::apply	(	const matrix< dpoint > &	src,
		dvector &	dest,
		const ivector &	indices,
		int	numCorrespondences
	)			const `[pure virtual]`

Estimates a transform from the supplied point sets, whereby only the points specified in the index vector are considered.

This method is used by robost estimators using a monte carlo approach.

All points of one point set give a matrix row, whereas all elements of a specifec correspondence stand in a matrix column.

Parameters:

	src	matrix<dpoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns.
	dest	dvector the estimated transform.
	indices	which rows of the src matrix have to be considered.
	numCorrespondences	the first numCorrespondences indices are considered.

Returns:: true if the transform could be estimated and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::apply	(	const matrix< dpoint > &	src,
		dvector &	dest,
		dvector &	error
	)			const `[pure virtual]`

Estimates a transform from the supplied point sets, where all points are considered.

Usually this method calls the apply without the residual first, and then computes the residual.

All points of one point set give a matrix row, whereas all elements of a specifec correspondence stand in a matrix column.

Parameters:

	src	matrix<dpoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns.
	dest	dvector the estimated transform.
	error	dvector containing the deviation of each point from the estimated transform. Usually this is the residual or elementwise squared residual.

Returns:: true if the transform could be estimated and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::apply	(	const matrix< dpoint > &	src,
		dvector &	dest
	)			const `[pure virtual]`

Estimates a transform from the supplied point sets, where all points are considered.

Please implement efficient code using iterators here. Not all robust estimators use a random sampling approach. Some estimators consider all input points and weight them according to their deviation from the transform computed at the prior iteration.

All points of one point set give a matrix row, whereas all elements of a specifec correspondence stand in a matrix column.

Parameters:

	src	matrix<dpoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns.
	dest	dvector the estimated transform.

Returns:: true if the transform could be estimated and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::apply	(	const matrix< fpoint > &	src,
		fvector &	dest,
		fvector &	error,
		const ivector &	indices,
		int	numCorrespondences
	)			const `[pure virtual]`

Estimates a transform from the supplied point sets, whereby only the points specified in the index vector are considered.

Usually this method calls the apply without the residual first, and then computes the residual.

All points of one point set give a matrix row, whereas all elements of a specifec correspondence stand in a matrix column.

Parameters:

	src	matrix<fpoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns.
	dest	fvector the estimated transform.
	error	fvector containing the deviation of each point from the estimated transform. Usually this is the residual or elementwise squared residual. All correspondences are considered, discarding the valid indices.
	indices	ivector with the indices of the relevant points.
	numCorrespondences	the first numCorrespondences indices are considered.

Returns:: true if the transform could be estimated and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::apply	(	const matrix< fpoint > &	src,
		fvector &	dest,
		const ivector &	indices,
		int	numCorrespondences
	)			const `[pure virtual]`

Estimates a transform from the supplied point sets, whereby only the points specified in the index vector are considered.

This method is used by robost estimators using a monte carlo approach.

All points of one point set give a matrix row, whereas all elements of a specifec correspondence stand in a matrix column.

Parameters:

	src	matrix<dpoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns.
	dest	fvector the estimated transform.
	indices	ivector with the indices of the relevant points.
	numCorrespondences	the first numCorrespondences indices are considered.

Returns:: true if the transform could be estimated and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::apply	(	const matrix< fpoint > &	src,
		fvector &	dest,
		fvector &	error
	)			const `[pure virtual]`

Estimates a transform from the supplied point sets, where all points are considered.

Usually this method calls the apply without the residual first, and then computes the residual.

Parameters:

	src	matrix<fpoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns. All points of one point set give a matrix row, whereas all elements of a specific correspondence stand in a matrix column.
	dest	fvector the estimated transform.
	error	fvector containing the deviation of each point from the estimated transform. Usually this is the residual or elementwise squared residual.

Returns:: true if the transform could be estimated and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::apply	(	const matrix< fpoint > &	src,
		fvector &	dest
	)			const `[pure virtual]`

Estimates a transform from the supplied point sets, where all points are considered.

Please implement efficient code using iterators here. Not all robust estimators use a random sampling approach. Some estimators consider all input points and weight them according to their deviation from the transform computed at the prior iteration.

All points of one point set give a matrix row, whereas all elements of a specifec correspondence stand in a matrix column.

Parameters:

	src	matrix<fpoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns.
	dest	fvector the estimated transform.

Returns:: true if the transform could be estimated and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::apply	(	const matrix< ipoint > &	src,
		fvector &	dest,
		fvector &	error,
		const ivector &	indices,
		int	numCorrespondences
	)			const `[pure virtual]`

Estimates a transform from the supplied point sets, whereby only the points specified in the index vector are considered.

Usually this method calls the apply without the residual first, and then computes the residual.

All points of one point set give a matrix row, whereas all elements of a specifec correspondence stand in a matrix column.

Parameters:

	src	matrix<ipoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns.
	dest	fvector the estimated transform.
	error	fvector containing the deviation of each point from the estimated transform. Usually this is the residual or elementwise squared residual. All correspondences are considered, discarding the valid indices.
	indices	ivector with the indices of the relevant points.
	numCorrespondences	the first numCorrespondences indices are considered.

Returns:: true if the transform could be estimated and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::apply	(	const matrix< ipoint > &	src,
		fvector &	dest,
		const ivector &	indices,
		int	numCorrespondences
	)			const `[pure virtual]`

Estimates a transform from the supplied point sets, whereby only the points specified in the index vector are considered.

This method is used by robost estimators using a monte carlo approach.

All points of one point set give a matrix row, whereas all elements of a specifec correspondence stand in a matrix column.

Parameters:

	src	matrix<ipoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns.
	dest	fvector the estimated transform.
	indices	ivector with the indices of the relevant points.
	numCorrespondences	the first numCorrespondences indices are considered.

Returns:: true if the transform could be estimated and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::apply	(	const matrix< ipoint > &	src,
		fvector &	dest,
		fvector &	error
	)			const `[pure virtual]`

Estimates a transform from the supplied point sets, where all points are considered.

Usually this method calls the apply without the residual first, and then computes the residual.

Parameters:

	src	matrix<ipoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns. All points of one point set give a matrix row, whereas all elements of a specific correspondence stand in a matrix column.
	dest	fvector the estimated transform.
	error	fvector containing the deviation of each point from the estimated transform. Usually this is the residual or elementwise squared residual.

Returns:: true if the transform could be estimated and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::apply	(	const matrix< ipoint > &	src,
		fvector &	dest
	)			const `[pure virtual]`

Estimates a transform from the supplied point sets, where all points are considered.

Please implement efficient code using iterators here. Not all robust estimators use a random sampling approach. Some estimators consider all input points and weight them according to their deviation from the transform computed at the prior iteration.

All points of one point set give a matrix row, whereas all elements of a specifec correspondence stand in a matrix column.

Parameters:

	src	matrix<ipoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns.
	dest	fvector the estimated transform.

Returns:: true if the transform could be estimated and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual functor* lti::transformEstimator::clone ( ) const [pure virtual]

Returns a pointer to a clone of this functor.

Implements lti::functor.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::computeResidual	(	const matrix< dpoint > &	src,
		const dvector &	transform,
		dvector &	dest
	)			const `[pure virtual]`

Compute the residual for the given correspondences and transformation vector.

Parameters:

	src	matrix<dpoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns.
	transform	dvector with the transformation
	dest	dvector with the residual

Returns:: true on success and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::computeResidual	(	const matrix< fpoint > &	src,
		const fvector &	transform,
		fvector &	dest
	)			const `[pure virtual]`

Compute the residual for the given correspondences and transformation vector.

Parameters:

	src	matrix<fpoint> with the point sets. All points of the same image stand in a row. The correspondences in another image stand in the according columns.
	transform	fvector with the transformation
	dest	fvector with the residual

Returns:: true on success and false otherwise.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

transformEstimator& lti::transformEstimator::copy ( const transformEstimator & other )

Copy data of "other" functor.

Parameters:

other

the functor to be copied

Returns:: a reference to this functor object

Reimplemented from lti::functor.

Reimplemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::denormalize	(	dvector &	srcdest,
		const vector< dpoint > &	scale,
		const vector< dpoint > &	shift
	)			const `[pure virtual]`

A transform estimated on normalized data usually differs from the transform of the original data.

Considering the normalization performed this methods computes the transform which applies to the original data.

Parameters:

	srcdest	the normalized transform. The result will be left here too.
	scale	a vector containing the scale applied to each point set.
	shift	a vector containing the shift of each scaled point set.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual bool lti::transformEstimator::denormalize	(	fvector &	srcdest,
		const vector< fpoint > &	scale,
		const vector< fpoint > &	shift
	)			const `[pure virtual]`

A transform estimated on normalized data usually differs from the transform of the original data.

Considering the normalization performed this methods computes the transform which applies to the original data.

Parameters:

	srcdest	the normalized transform. The result will be left here too.
	scale	a vector containing the scale applied to each point set.
	shift	a vector containing the shift of each scaled point set.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

const parameters& lti::transformEstimator::getParameters ( ) const

Returns used parameters.

Reimplemented from lti::functor.

Reimplemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual const char* lti::transformEstimator::getTypeName ( ) const [virtual]

Returns the name of this type ("transformEstimator").

Reimplemented from lti::functor.

Reimplemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual int lti::transformEstimator::maxCorrespondenceDimension ( ) const [pure virtual]

Returns the maximum dimension of a correspondence, e.g.

the maximum dimension of a correspondence pair is 2, whereas transformEstimator running on n-tuples may allow an infinite number. Each derived transform estimator only works on correspondences of priori defined dimensions.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual int lti::transformEstimator::minCorrespondenceDimension ( ) const [pure virtual]

Returns the mininum dimension of a correspondence, e.g.

the minimum dimension of a correspondence pair is 2. Each derived transform estimator only works on correspondences of priori defined dimensions.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

virtual int lti::transformEstimator::minNumberCorrespondences ( ) const [pure virtual]

Returns the minimum number of correspondences required to estimate the transform.

Implemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

transformEstimator& lti::transformEstimator::operator= ( const transformEstimator & other )

Alias for copy member.

Parameters:

other

the functor to be copied

Returns:: a reference to this functor object

Reimplemented from lti::functor.

Reimplemented in lti::fMatrixEstimator, lti::fMatrixEstimatorBase, lti::homography8DofEstimator, lti::homography9DofEstimator, lti::homographyEstimatorBase, and lti::translationScaleEstimator.

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

ltiTransformEstimator.h

lti::transformEstimator Class Reference

Classes

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation