Ocean::Tracking::HomographyImageAlignmentDense Class Reference

This class implements functions allowing dense image alignment. More...

Data Structures
class	ConsistencyData
	Definition of an abstract base class allowing to store consistency data. More...
class	DenseAdditiveHomographyProvider
class	DenseInverseCompositionalHomographyProvider
class	HomographyProvider

Public Types
typedef ObjectRef< ConsistencyData >	ConsistencyDataRef
	Definition of an object reference holding a consistency data object. More...

Static Public Member Functions
static bool	optimizeAlignmentAdditive (const Frame &templateFrame, const CV::SubRegion &templateSubRegion, const Frame &currentFrame, const SquareMatrix3 &roughHomography, const unsigned int homographyParameters, const bool zeroMean, SquareMatrix3 &homography, const unsigned int iterations=20u, Scalar lambda=10, const Scalar lambdaFactor=10, Scalar *initialError=nullptr, Scalar *finalError=nullptr, Scalars *intermediateErrors=nullptr, ConsistencyDataRef *externalConsistencyData=nullptr, bool *abort=nullptr)
	Optimizes the alignment between two images within a specified sub-region regarding a homography by application of an iterative additive delta Levenberg-Marquardt optimization approach. More...
static bool	optimizeAlignmentInverseCompositional (const Frame &templateFrame, const CV::SubRegion &templateSubRegion, const Frame &currentFrame, const SquareMatrix3 &roughHomography, const unsigned int homographyParameters, const bool zeroMean, SquareMatrix3 &homography, const unsigned int iterations=20u, Scalar lambda=10, const Scalar lambdaFactor=10, Scalar *initialError=nullptr, Scalar *finalError=nullptr, Scalars *intermediateErrors=nullptr, ConsistencyDataRef *externalConsistencyData=nullptr, bool *abort=nullptr)
	Optimizes the alignment between two images within a specified sub-region regarding a homography by application of an iterative inverse compositional delta Levenberg-Marquardt optimization approach. More...
static bool	optimizeAlignmentMultiResolution (const Frame &templateFrame, const CV::SubRegion &templateSubRegion, const Frame &currentFrame, const unsigned int numberPyramidLayers, const unsigned int homographyParameters, const bool additiveAlignment, const bool levenbergMarquardtOptimization, const bool zeroMean, const SquareMatrix3 &roughHomography, SquareMatrix3 &homography, const unsigned int coarseIterations=20u, const unsigned int fineIterations=4u, const CV::FramePyramid::DownsamplingMode downsamplingMode=CV::FramePyramid::DM_FILTER_14641)
	Optimizes the alignment between two images within a specified sub-region regarding a homography by applying a multi-resolution (coarse to fine approach) Gauss-Newton or a Levenberg-Marquardt optimization approach. More...
static bool	optimizeAlignmentMultiResolution (const CV::FramePyramid &templateFramePyramid, const std::vector< CV::SubRegion > &templateSubRegions, const CV::FramePyramid &currentFramePyramid, const unsigned int layers, const unsigned int homographyParameters, const bool additiveAlignment, const bool levenbergMarquardtOptimization, const bool zeroMean, const SquareMatrix3 &roughHomography, SquareMatrix3 &homography, const unsigned int coarseIterations=20u, const unsigned int fineIterations=4u, std::vector< Tracking::HomographyImageAlignmentDense::ConsistencyDataRef > *consistencyDatas=nullptr)
	Optimizes the alignment between two images within a specified sub-region regarding a homography by applying a multi-resolution (coarse to fine approach) Gauss-Newton or a Levenberg-Marquardt optimization approach. More...

Static Protected Member Functions
static Scalar	slowDetermineError (const Frame &templateFrame, const CV::SubRegion &templateSubRegion, const Frame &currentFrame, const SquareMatrix3 &homography, const bool zeroMean)
	Determines the current error for a given homography between a current frame and a template frame within a sub-region of the template frame. More...
template<unsigned int tChannels>
static Scalar	slowDetermineError8BitPerChannel (const Frame &templateFrame, const CV::SubRegion &templateSubRegion, const Frame &currentFrame, const SquareMatrix3 &homography, const bool zeroMean)
	Determines the current error for a given homography between a current frame and a template frame within a sub-region of the template frame. More...
template<unsigned int tParameters>
static bool	slowDetermineHessianAndErrorJacobian (const Frame &templateFrame, const CV::SubRegion &templateSubRegion, const Frame &currentFrame, const SquareMatrix3 &homography, const bool zeroMean, Matrix &hessian, Matrix &jacobianError)
	Determines the 8x8 or 9x9 Hessian matrix and the 8x1 or 9x1 Jacobian-Error vector for a given homography between a current frame and a template frame within a sub-region of the template frame. More...
template<unsigned int tParameters, unsigned int tChannels>
static bool	slowDetermineHessianAndErrorJacobian8BitPerChannel (const Frame &templateFrame, const CV::SubRegion &templateSubRegion, const Frame &currentFrame, const SquareMatrix3 &homography, const bool zeroMean, Matrix &hessian, Matrix &jacobianError)
	Determines the 8x8 or 9x9 Hessian matrix and the 8x1 or 9x1 Jacobian-Error vector for a given homography between a current frame and a template frame within a sub-region of the template frame. More...
template<unsigned int tChannels, bool tDetermineTemplateMeans>
static bool	determineMeans8BitPerChannel (const Frame &templateFrame, const CV::SubRegion &templateSubRegion, const Frame &transformedCurrentSubFrame, const Frame &transformedCurrentSubFrameMask, Scalar *templateMeans, Scalar *currentMeans, Worker *worker)
	Determines the mean color intensities in the corresponding sub-regions of the current transformed frame and also optional in the template frame while optional a mask frame defines valid and invalid pixels (not counting for error determination). More...
template<unsigned int tChannels, bool tUseMeans>
static Scalar	determineError8BitPerChannel (const Frame &templateFrame, const CV::SubRegion &templateSubRegion, const Frame &transformedCurrentSubFrameData, const Frame &transformedCurrentSubFrameMaskData, const Scalar *templateMeans, const Scalar *currentMeans, Worker *worker)
	Determines the color intensity error within a sub-region between a template frame and a transformed current frame while optional a mask frame defines valid and invalid pixels (not counting for error determination). More...
template<unsigned int tParameters, unsigned int tChannels, bool tUseMeans>
static void	determineHessianAndErrorJacobian8BitPerChannel (const Frame &templateFrame, const CV::SubRegion &templateSubRegion, const Frame &transformedCurrentSubFrame, const Frame &transformedCurrentSubFrameMask, const unsigned int transformedBoundingBoxLeft, const unsigned int transformedBoundingBoxTop, const unsigned int transformedBoundingBoxWidth, const unsigned int transformedBoundingBoxHeight, const Frame &gradientCurrentFrame, const SquareMatrix3 &homography, const Scalar *templateMeans, const Scalar *currentMeans, Matrix &hessian, Matrix &jacobianError, Worker *worker)
	Determines the Hessian matrix and the product of transposed Jacobian matrix and error vector for a template frame and current transformed frame within a sub-region while optional a mask frame defines valid and invalid pixels (not counting for error determination). More...
template<unsigned int tParameters, unsigned int tChannels, bool tUseMeans>
static void	determineErrorJacobianInverseCompositional8BitPerChannel (const Frame &templateFrame, const CV::SubRegion &templateSubRegion, const Frame &transformedCurrentSubFrame, const Frame &transformedCurrentSubFrameMask, const Scalar *templateMeans, const Scalar *currentMeans, const Scalar *jacobianRows, Matrix &jacobianError, Worker *worker)
	Determines the product of transposed Jacobian matrix and error vector for the inverse compositional optimization approach for a template frame and current transformed frame within a sub-region while optional a mask frame defines valid and invalid pixels (not counting for error determination). More...
template<unsigned int tChannels, bool tDetermineTemplateMeans>
static void	determineMeans8BitPerChannelSubset (const Frame *templateFrame, const CV::SubRegion *templateSubRegion, const Frame *transformedCurrentSubFrame, Scalar *templateMeans, Scalar *currentMeans, unsigned int *templateMeansDenominator, unsigned int *currentMeansDenominator, Lock *lock, const unsigned int threads, const unsigned int threadIndex, const unsigned int unused)
	Determines the mean color intensities in a subset of the corresponding sub-regions of the current transformed frame and also optional in the template frame. More...
template<unsigned int tChannels, bool tDetermineTemplateMeans>
static void	determineMeansMask8BitPerChannelSubset (const Frame *templateFrame, const CV::SubRegion *templateSubRegion, const Frame *transformedCurrentSubFrame, const Frame *transformedCurrentSubFrameMask, Scalar *templateMeans, Scalar *currentMeans, unsigned int *templateMeansDenominator, unsigned int *currentMeansDenominator, Lock *lock, const unsigned int threads, const unsigned int threadIndex, const unsigned int unused)
	Determines the mean color intensities in a subset of the corresponding sub-regions of the current transformed frame and also optional in the template frame. More...
template<unsigned int tChannels, bool tUseMeans>
static void	determineError8BitPerChannelSubset (const Frame *templateFrame, const CV::SubRegion *templateSubRegion, const Frame *transformedCurrentSubFrame, const Scalar *templateMeans, const Scalar *currentMeans, Scalar *squaredError, unsigned int *errorDenominators, Lock *lock, const unsigned int threads, const unsigned int threadIndex, const unsigned int unused)
	Determines the color intensity error in a subset of a sub-region between a template frame and a transformed current frame. More...
template<unsigned int tChannels, bool tUseMeans>
static void	determineErrorMask8BitPerChannelSubset (const Frame *templateFrame, const CV::SubRegion *templateSubRegion, const Frame *transformedCurrentSubFrame, const Frame *transformedCurrentSubFrameMask, const Scalar *templateMeans, const Scalar *currentMeans, Scalar *squaredError, unsigned int *errorDenominators, Lock *lock, const unsigned int threads, const unsigned int threadIndex, const unsigned int unused)
	Determines the color intensity error in a subset of a sub-region between a template frame and a transformed current frame while a mask frame defines valid and invalid pixels (not counting for error determination). More...
template<unsigned int tParameters, unsigned int tChannels, bool tUseMeans>
static void	determineHessianAndErrorJacobian8BitPerChannelSubset (const Frame *templateFrame, const CV::SubRegion *templateSubRegion, const Frame *transformedCurrentSubFrame, const unsigned int transformedBoundingBoxLeft, const unsigned int transformedBoundingBoxTop, const unsigned int transformedBoundingBoxWidth, const unsigned int transformedBoundingBoxHeight, const Frame *gradientCurrentFrame, const SquareMatrix3 *homography, const Scalar *templateMeans, const Scalar *currentMeans, Matrix *hessian, Matrix *jacobianError, Lock *lock, const unsigned int threads, const unsigned int threadIndex, const unsigned int unused)
	Determines the subset-Hessian matrix and the subset-product of transposed Jacobian matrix and error vector for a template frame and current transformed frame within a sub-region. More...
template<unsigned int tParameters, unsigned int tChannels, bool tUseMeans>
static void	determineErrorJacobianInverseCompositional8BitPerChannelSubset (const Frame *templateFrame, const CV::SubRegion *templateSubRegion, const Frame *transformedCurrentSubFrame, const Scalar *templateMeans, const Scalar *currentMeans, const Scalar *jacobianRows, Matrix *jacobianError, Lock *lock, const unsigned int threads, const unsigned int threadIndex, const unsigned int unused)
	Determines the subset of the product of transposed Jacobian matrix and error vector for the inverse compositional optimization approach for a template frame and current transformed frame within a sub-region while optional a mask frame defines valid and invalid pixels (not counting for error determination). More...
template<unsigned int tParameters, unsigned int tChannels, bool tUseMeans>
static void	determineHessianAndErrorJacobianMask8BitPerChannelSubset (const Frame *templateFrame, const CV::SubRegion *templateSubRegion, const Frame *transformedCurrentSubFrame, const Frame *transformedCurrentSubFrameMask, const unsigned int transformedBoundingBoxLeft, const unsigned int transformedBoundingBoxTop, const unsigned int transformedBoundingBoxWidth, const unsigned int transformedBoundingBoxHeight, const Frame *gradientCurrentFrame, const SquareMatrix3 *current_H_template, const Scalar *templateMeans, const Scalar *currentMeans, Matrix *hessian, Matrix *jacobianError, Lock *lock, const unsigned int threads, const unsigned int threadIndex, const unsigned int unused)
	Determines the subset-Hessian matrix and the subset-product of transposed Jacobian matrix and error vector for a template frame and current transformed frame within a sub-region. More...

Detailed Description

This class implements functions allowing dense image alignment.

Member Typedef Documentation

ConsistencyDataRef

typedef ObjectRef<ConsistencyData> Ocean::Tracking::HomographyImageAlignmentDense::ConsistencyDataRef

Definition of an object reference holding a consistency data object.

Member Function Documentation

determineError8BitPerChannel()

template<unsigned int tChannels, bool tUseMeans>

Scalar Ocean::Tracking::HomographyImageAlignmentDense::determineError8BitPerChannel ( const Frame &  templateFrame, const CV::SubRegion &  templateSubRegion, const Frame &  transformedCurrentSubFrameData, const Frame &  transformedCurrentSubFrameMaskData, const Scalar *  templateMeans, const Scalar *  currentMeans, Worker *  worker  )

inlinestaticprotected

Determines the color intensity error within a sub-region between a template frame and a transformed current frame while optional a mask frame defines valid and invalid pixels (not counting for error determination).

The homography transformed points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint).

Parameters

templateFrame	The frame data of the template frame, must be valid
templateSubRegion	The sub-region for which the error is determined, defined in the template frame
transformedCurrentSubFrameData	The sub-region from the transformed current frame, needs to be transformed with the homography with size and location equal to the sub-region
transformedCurrentSubFrameMaskData	Optional mask that has been created during the creation of the sub-region of the transformed current frame, nullptr to avoid the usage of a mask which is more efficient
templateMeans	The mean color intensities in the sub-region of the template frame, must be defined if 'tUseMeans' is True; otherwise nullptr
currentMeans	The mean color intensities in the current transformed frame, must be defined if 'tUseMeans' is True; otherwise nullptr
worker	Optional worker object to distribute the computation

Returns

The average squared error for each pixel in the sub-region

Template Parameters

tChannels	The number of data channels of the provided frames
tUseMeans	True, if the mean color intensities are used to determine the errors; False, otherwise

determineError8BitPerChannelSubset()

template<unsigned int tChannels, bool tUseMeans>

static void Ocean::Tracking::HomographyImageAlignmentDense::determineError8BitPerChannelSubset ( const Frame *  templateFrame, const CV::SubRegion *  templateSubRegion, const Frame *  transformedCurrentSubFrame, const Scalar *  templateMeans, const Scalar *  currentMeans, Scalar *  squaredError, unsigned int *  errorDenominators, Lock *  lock, const unsigned int  threads, const unsigned int  threadIndex, const unsigned int  unused  )

staticprotected

Determines the color intensity error in a subset of a sub-region between a template frame and a transformed current frame.

The homography transformed points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint).

Parameters

templateFrame	The template frame, must be valid
templateSubRegion	The sub-region for which the error is determined, defined in the template frame
transformedCurrentSubFrame	The sub-region from the transformed current frame, needs to be transformed with the homography with size and location equal to the sub-region
templateMeans	The mean color intensities in the sub-region of the template frame, must be defined if 'tUseMeans' is True; otherwise nullptr
currentMeans	The mean color intensities in the current transformed frame, must be defined if 'tUseMeans' is True; otherwise nullptr
squaredError	The resulting sum of squared errors, must be initialized with 0
errorDenominators	The number of pixels that have been used to determine the error, must be initialized with 0
lock	The lock object necessary if this function is executed on multiple threads in parallel, nullptr otherwise
threads	The number of threads on which this function is executed in parallel
threadIndex	The index of the current thread, with range [0, threads)
unused	An unused parameter must be 1

Template Parameters

tChannels	The number of data channels of the provided frames
tUseMeans	True, if the mean color intensities are used to determine the errors; False, otherwise

determineErrorJacobianInverseCompositional8BitPerChannel()

template<unsigned int tParameters, unsigned int tChannels, bool tUseMeans>

void Ocean::Tracking::HomographyImageAlignmentDense::determineErrorJacobianInverseCompositional8BitPerChannel ( const Frame &  templateFrame, const CV::SubRegion &  templateSubRegion, const Frame &  transformedCurrentSubFrame, const Frame &  transformedCurrentSubFrameMask, const Scalar *  templateMeans, const Scalar *  currentMeans, const Scalar *  jacobianRows, Matrix &  jacobianError, Worker *  worker  )

inlinestaticprotected

Determines the product of transposed Jacobian matrix and error vector for the inverse compositional optimization approach for a template frame and current transformed frame within a sub-region while optional a mask frame defines valid and invalid pixels (not counting for error determination).

Parameters

templateFrame	The template frame, must be valid
templateSubRegion	The sub-region for which the error is determined, defined in the template frame
transformedCurrentSubFrame	The sub-region from the transformed current frame, needs to be transformed with the homography with size and location equal to the sub-region
transformedCurrentSubFrameMask	Optional mask that has been created during the creation of the sub-region of the transformed current frame, nullptr to avoid the usage of any mask which is more efficient
templateMeans	The mean color intensities in the sub-region of the template frame, must be defined if 'tUseMeans' is True; otherwise nullptr
currentMeans	The mean color intensities in the current transformed frame, must be defined if 'tUseMeans' is True; otherwise nullptr
jacobianRows	The already determine (and constant) Jacobian rows for each pixel and channel of the template sub-region, each row has dimension (1 x tParameters)
jacobianError	The resulting product of transposed Jacobian matrix and error vector, with dimension (tParameters x 1)
worker	Optional worker object to distribute the computation

Template Parameters

tParameters	The number of parameters that are used to optimize the homography, with range [8, 9]
tChannels	The number of data channels of the provided frames
tUseMeans	True, if the mean color intensities are used to determine the errors; False, otherwise

determineErrorJacobianInverseCompositional8BitPerChannelSubset()

template<unsigned int tParameters, unsigned int tChannels, bool tUseMeans>

static void Ocean::Tracking::HomographyImageAlignmentDense::determineErrorJacobianInverseCompositional8BitPerChannelSubset ( const Frame *  templateFrame, const CV::SubRegion *  templateSubRegion, const Frame *  transformedCurrentSubFrame, const Scalar *  templateMeans, const Scalar *  currentMeans, const Scalar *  jacobianRows, Matrix *  jacobianError, Lock *  lock, const unsigned int  threads, const unsigned int  threadIndex, const unsigned int  unused  )

staticprotected

Determines the subset of the product of transposed Jacobian matrix and error vector for the inverse compositional optimization approach for a template frame and current transformed frame within a sub-region while optional a mask frame defines valid and invalid pixels (not counting for error determination).

Parameters

templateFrame	The template frame, must be valid
templateSubRegion	The sub-region for which the error is determined, defined in the template frame
transformedCurrentSubFrame	The sub-region from the transformed current frame, needs to be transformed with the homography with size and location equal to the sub-region
templateMeans	The mean color intensities in the sub-region of the template frame, must be defined if 'tUseMeans' is True; otherwise nullptr
currentMeans	The mean color intensities in the current transformed frame, must be defined if 'tUseMeans' is True; otherwise nullptr
jacobianRows	The already determine (and constant) Jacobian rows for each pixel and channel of the template sub-region, each row has dimension (1 x tParameters)
jacobianError	The resulting product of transposed Jacobian matrix and error vector, with dimension (tParameters x 1), must be initialized with 0
lock	The lock object necessary if this function is executed on multiple threads in parallel, nullptr otherwise
threads	The number of threads on which this function is executed in parallel
threadIndex	The index of the current thread, with range [0, threads)
unused	An unused parameter must be 1

Template Parameters

tParameters	The number of parameters that are used to optimize the homography, with range [8, 9]
tChannels	The number of data channels of the provided frames
tUseMeans	True, if the mean color intensities are used to determine the errors; False, otherwise

determineErrorMask8BitPerChannelSubset()

template<unsigned int tChannels, bool tUseMeans>

static void Ocean::Tracking::HomographyImageAlignmentDense::determineErrorMask8BitPerChannelSubset ( const Frame *  templateFrame, const CV::SubRegion *  templateSubRegion, const Frame *  transformedCurrentSubFrame, const Frame *  transformedCurrentSubFrameMask, const Scalar *  templateMeans, const Scalar *  currentMeans, Scalar *  squaredError, unsigned int *  errorDenominators, Lock *  lock, const unsigned int  threads, const unsigned int  threadIndex, const unsigned int  unused  )

staticprotected

Determines the color intensity error in a subset of a sub-region between a template frame and a transformed current frame while a mask frame defines valid and invalid pixels (not counting for error determination).

The homography transformed points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint).

Parameters

templateFrame	The template frame, must be valid
templateSubRegion	The sub-region for which the error is determined, defined in the template frame
transformedCurrentSubFrame	The sub-region from the transformed current frame, needs to be transformed with the homography with size and location equal to the sub-region
transformedCurrentSubFrameMask	The mask that has been created during the creation of the sub-region of the transformed current frame
templateMeans	The mean color intensities in the sub-region of the template frame, must be defined if 'tUseMeans' is True; otherwise nullptr
currentMeans	The mean color intensities in the current transformed frame, must be defined if 'tUseMeans' is True; otherwise nullptr
squaredError	The resulting sum of squared errors, must be initialized with 0
errorDenominators	The number of pixels that have been used to determine the error, must be initialized with 0
lock	The lock object necessary if this function is executed on multiple threads in parallel, nullptr otherwise
threads	The number of threads on which this function is executed in parallel
threadIndex	The index of the current thread, with range [0, threads)
unused	An unused parameter must be 1

Template Parameters

tChannels	The number of data channels of the provided frames
tUseMeans	True, if the mean color intensities are used to determine the errors; False, otherwise

determineHessianAndErrorJacobian8BitPerChannel()

template<unsigned int tParameters, unsigned int tChannels, bool tUseMeans>

void Ocean::Tracking::HomographyImageAlignmentDense::determineHessianAndErrorJacobian8BitPerChannel ( const Frame &  templateFrame, const CV::SubRegion &  templateSubRegion, const Frame &  transformedCurrentSubFrame, const Frame &  transformedCurrentSubFrameMask, const unsigned int  transformedBoundingBoxLeft, const unsigned int  transformedBoundingBoxTop, const unsigned int  transformedBoundingBoxWidth, const unsigned int  transformedBoundingBoxHeight, const Frame &  gradientCurrentFrame, const SquareMatrix3 &  homography, const Scalar *  templateMeans, const Scalar *  currentMeans, Matrix &  hessian, Matrix &  jacobianError, Worker *  worker  )

inlinestaticprotected

Determines the Hessian matrix and the product of transposed Jacobian matrix and error vector for a template frame and current transformed frame within a sub-region while optional a mask frame defines valid and invalid pixels (not counting for error determination).

The homography transformed points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint).

Parameters

templateFrame	The template frame, must be valid
templateSubRegion	The sub-region for which the error is determined, defined in the template frame
transformedCurrentSubFrame	The sub-region from the transformed current frame, needs to be transformed with the homography with size and location equal to the sub-region
transformedCurrentSubFrameMask	Optional mask that has been created during the creation of the sub-region of the transformed current frame, nullptr to avoid the usage of any mask which is more efficient
transformedBoundingBoxLeft	The left border of the bounding box of the transformed sub-region in pixel, with range [0, currentFrame.width())
transformedBoundingBoxTop	the top border of the bounding box of the transformed sub-region in pixel, with range [0, currentFrame.height())
transformedBoundingBoxWidth	The width of the bounding box of the transformed sub-region in pixel, with range [1, currentFrame.width() - transformedBoundingBoxLeft]
transformedBoundingBoxHeight	The height of the bounding box of the transformed sub-region in pixel, with range [1, currentFrame.height() - transformedBoundingBoxTop]
gradientCurrentFrame	The gradient filter responses of the current frame restricted to the location and size of the transformed bounding box, must be valid
homography	The homography to be used to transform the current frame with
templateMeans	The mean color intensities in the sub-region of the template frame, must be defined if 'tUseMeans' is True; otherwise nullptr
currentMeans	The mean color intensities in the current transformed frame, must be defined if 'tUseMeans' is True; otherwise nullptr
hessian	The resulting Hessian matrix, with dimension (tParameters x tParameters)
jacobianError	The resulting product of transposed Jacobian matrix and error vector, with dimension (tParameters x 1)
worker	Optional worker object to distribute the computation

Template Parameters

tParameters	The number of parameters that are used to optimize the homography, with range [8, 9]
tChannels	The number of data channels of the provided frames
tUseMeans	True, if the mean color intensities are used to determine the errors; False, otherwise

determineHessianAndErrorJacobian8BitPerChannelSubset()

template<unsigned int tParameters, unsigned int tChannels, bool tUseMeans>

static void Ocean::Tracking::HomographyImageAlignmentDense::determineHessianAndErrorJacobian8BitPerChannelSubset ( const Frame *  templateFrame, const CV::SubRegion *  templateSubRegion, const Frame *  transformedCurrentSubFrame, const unsigned int  transformedBoundingBoxLeft, const unsigned int  transformedBoundingBoxTop, const unsigned int  transformedBoundingBoxWidth, const unsigned int  transformedBoundingBoxHeight, const Frame *  gradientCurrentFrame, const SquareMatrix3 *  homography, const Scalar *  templateMeans, const Scalar *  currentMeans, Matrix *  hessian, Matrix *  jacobianError, Lock *  lock, const unsigned int  threads, const unsigned int  threadIndex, const unsigned int  unused  )

staticprotected

Determines the subset-Hessian matrix and the subset-product of transposed Jacobian matrix and error vector for a template frame and current transformed frame within a sub-region.

The homography transformed points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint).

Parameters

templateFrame	The template frame, must be valid
templateSubRegion	The sub-region for which the error is determined, defined in the template frame
transformedCurrentSubFrame	The sub-region from the transformed current frame, needs to be transformed with the homography with size and location equal to the sub-region
transformedBoundingBoxLeft	The left border of the bounding box of the transformed sub-region in pixel, with range [0, currentFrame.width())
transformedBoundingBoxTop	the top border of the bounding box of the transformed sub-region in pixel, with range [0, currentFrame.height())
transformedBoundingBoxWidth	The width of the bounding box of the transformed sub-region in pixel, with range [1, currentFrame.width() - transformedBoundingBoxLeft]
transformedBoundingBoxHeight	The height of the bounding box of the transformed sub-region in pixel, with range [1, currentFrame.height() - transformedBoundingBoxTop]
gradientCurrentFrame	The gradient filter responses of the current frame restricted to the location and size of the transformed bounding box, must be valid
homography	The homography to be used to transform the current frame with
templateMeans	The mean color intensities in the sub-region of the template frame, must be defined if 'tUseMeans' is True; otherwise nullptr
currentMeans	The mean color intensities in the current transformed frame, must be defined if 'tUseMeans' is True; otherwise nullptr
hessian	The resulting Hessian matrix, with dimension (tParameters x tParameters), must be initialized with 0
jacobianError	The resulting product of transposed Jacobian matrix and error vector, with dimension (tParameters x 1), must be initialized with 0
lock	The lock object necessary if this function is executed on multiple threads in parallel, nullptr otherwise
threads	The number of threads on which this function is executed in parallel
threadIndex	The index of the current thread, with range [0, threads)
unused	An unused parameter must be 1

Template Parameters

tParameters	The number of parameters that are used to optimize the homography, with range [8, 9]
tChannels	The number of data channels of the provided frames
tUseMeans	True, if the mean color intensities are used to determine the errors; False, otherwise

determineHessianAndErrorJacobianMask8BitPerChannelSubset()

template<unsigned int tParameters, unsigned int tChannels, bool tUseMeans>

static void Ocean::Tracking::HomographyImageAlignmentDense::determineHessianAndErrorJacobianMask8BitPerChannelSubset ( const Frame *  templateFrame, const CV::SubRegion *  templateSubRegion, const Frame *  transformedCurrentSubFrame, const Frame *  transformedCurrentSubFrameMask, const unsigned int  transformedBoundingBoxLeft, const unsigned int  transformedBoundingBoxTop, const unsigned int  transformedBoundingBoxWidth, const unsigned int  transformedBoundingBoxHeight, const Frame *  gradientCurrentFrame, const SquareMatrix3 *  current_H_template, const Scalar *  templateMeans, const Scalar *  currentMeans, Matrix *  hessian, Matrix *  jacobianError, Lock *  lock, const unsigned int  threads, const unsigned int  threadIndex, const unsigned int  unused  )

staticprotected

Determines the subset-Hessian matrix and the subset-product of transposed Jacobian matrix and error vector for a template frame and current transformed frame within a sub-region.

Parameters

templateFrame	The template frame, must be valid
templateSubRegion	The sub-region for which the error is determined, defined in the template frame
transformedCurrentSubFrame	The sub-region from the transformed current frame, needs to be transformed with the homography with size and location equal to the sub-region
transformedCurrentSubFrameMask	The mask that has been created during the creation of the sub-region of the transformed current frame
transformedBoundingBoxLeft	The left border of the bounding box of the transformed sub-region in pixel, with range [0, currentFrame.width())
transformedBoundingBoxTop	the top border of the bounding box of the transformed sub-region in pixel, with range [0, currentFrame.height())
transformedBoundingBoxWidth	The width of the bounding box of the transformed sub-region in pixel, with range [1, currentFrame.width() - transformedBoundingBoxLeft]
transformedBoundingBoxHeight	The height of the bounding box of the transformed sub-region in pixel, with range [1, currentFrame.height() - transformedBoundingBoxTop]
gradientCurrentFrame	The gradient filter responses of the current frame restricted to the location and size of the transformed bounding box, must be valid
current_H_template	The homography converting the template image to the current image, must be valid
templateMeans	The mean color intensities in the sub-region of the template frame, must be defined if 'tUseMeans' is True; otherwise nullptr
currentMeans	The mean color intensities in the current transformed frame, must be defined if 'tUseMeans' is True; otherwise nullptr
hessian	The resulting Hessian matrix, with dimension (tParameters x tParameters), must be initialized with 0
jacobianError	The resulting product of transposed Jacobian matrix and error vector, with dimension (tParameters x 1), must be initialized with 0
lock	The lock object necessary if this function is executed on multiple threads in parallel, nullptr otherwise
threads	The number of threads on which this function is executed in parallel
threadIndex	The index of the current thread, with range [0, threads)
unused	An unused parameter must be 1

Template Parameters

tParameters	The number of parameters that are used to optimize the homography, with range [8, 9]
tChannels	The number of data channels of the provided frames
tUseMeans	True, if the mean color intensities are used to determine the errors; False, otherwise

determineMeans8BitPerChannel()

template<unsigned int tChannels, bool tDetermineTemplateMeans>

bool Ocean::Tracking::HomographyImageAlignmentDense::determineMeans8BitPerChannel ( const Frame &  templateFrame, const CV::SubRegion &  templateSubRegion, const Frame &  transformedCurrentSubFrame, const Frame &  transformedCurrentSubFrameMask, Scalar *  templateMeans, Scalar *  currentMeans, Worker *  worker  )

inlinestaticprotected

Determines the mean color intensities in the corresponding sub-regions of the current transformed frame and also optional in the template frame while optional a mask frame defines valid and invalid pixels (not counting for error determination).

The homography transformed points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint).

Parameters

templateFrame	The template frame, must be valid
templateSubRegion	The sub-region in which the mean color intensities are determined, defined in the template frame
transformedCurrentSubFrame	The sub-region from the transformed current frame, needs to be transformed with the homography with size and location equal to the sub-region
transformedCurrentSubFrameMask	Optional mask that has been created during the creation of the sub-region of the transformed current frame, nullptr to avoid the usage of any mask which is more efficient
templateMeans	The resulting mean color intensities in the template frame, if 'tDetermineTemplateMeans' is True
currentMeans	The resulting mean color intensities in the current transformed frame
worker	Optional worker object to distribute the computation

Returns

True, if succeeded

Template Parameters

tChannels	The number of data channels of the provided frames
tDetermineTemplateMeans	True, to determine the mean intensities for the current frame and the template frame; False, to determine the mean intensities for the current frame only

determineMeans8BitPerChannelSubset()

template<unsigned int tChannels, bool tDetermineTemplateMeans>

static void Ocean::Tracking::HomographyImageAlignmentDense::determineMeans8BitPerChannelSubset ( const Frame *  templateFrame, const CV::SubRegion *  templateSubRegion, const Frame *  transformedCurrentSubFrame, Scalar *  templateMeans, Scalar *  currentMeans, unsigned int *  templateMeansDenominator, unsigned int *  currentMeansDenominator, Lock *  lock, const unsigned int  threads, const unsigned int  threadIndex, const unsigned int  unused  )

staticprotected

Determines the mean color intensities in a subset of the corresponding sub-regions of the current transformed frame and also optional in the template frame.

The homography transformed points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint).

Parameters

templateFrame	The template frame, must be valid
templateSubRegion	The sub-region in which the mean color intensities are determined, defined in the template frame
transformedCurrentSubFrame	The sub-region from the transformed current frame, needs to be transformed with the homography with size and location equal to the sub-region
templateMeans	The resulting sum of color intensities in the template frame, if 'tDetermineTemplateMeans' is True, must be initialized with 0
currentMeans	The resulting sum of color intensities in the current transformed frame, must be initialized with 0
templateMeansDenominator	The number of pixels that have been used to determine the sum of intensities in the template frame, must be initialized with 0
currentMeansDenominator	The number of pixels that have been used to determine the sum of intensities in the current transformed frame, must be initialized with 0
lock	The lock object necessary if this function is executed on multiple threads in parallel, nullptr otherwise
threads	The number of threads on which this function is executed in parallel
threadIndex	The index of the current thread, with range [0, threads)
unused	An unused parameter must be 1

Template Parameters

tChannels	The number of data channels of the provided frames
tDetermineTemplateMeans	True, to determine the mean intensities for the current frame and the template frame; False, to determine the mean intensities for the current frame only

See also

determineMeans8BitPerChannel().

determineMeansMask8BitPerChannelSubset()

template<unsigned int tChannels, bool tDetermineTemplateMeans>

static void Ocean::Tracking::HomographyImageAlignmentDense::determineMeansMask8BitPerChannelSubset ( const Frame *  templateFrame, const CV::SubRegion *  templateSubRegion, const Frame *  transformedCurrentSubFrame, const Frame *  transformedCurrentSubFrameMask, Scalar *  templateMeans, Scalar *  currentMeans, unsigned int *  templateMeansDenominator, unsigned int *  currentMeansDenominator, Lock *  lock, const unsigned int  threads, const unsigned int  threadIndex, const unsigned int  unused  )

staticprotected

Determines the mean color intensities in a subset of the corresponding sub-regions of the current transformed frame and also optional in the template frame.

The homography transformed points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint).

Parameters

templateFrame	The template frame, must be valid
templateSubRegion	The sub-region in which the mean color intensities are determined, defined in the template frame
transformedCurrentSubFrame	The sub-region from the transformed current frame, needs to be transformed with the homography with size and location equal to the sub-region
transformedCurrentSubFrameMask	The mask that has been created during the creation of the sub-region of the transformed current frame
templateMeans	The resulting sum of color intensities in the template frame, if 'tDetermineTemplateMeans' is True, must be initialized with 0
currentMeans	The resulting sum of color intensities in the current transformed frame, must be initialized with 0
templateMeansDenominator	The number of pixels that have been used to determine the sum of intensities in the template frame, must be initialized with 0
currentMeansDenominator	The number of pixels that have been used to determine the sum of intensities in the current transformed frame, must be initialized with 0
lock	The lock object necessary if this function is executed on multiple threads in parallel, nullptr otherwise
threads	The number of threads on which this function is executed in parallel
threadIndex	The index of the current thread, with range [0, threads)
unused	An unused parameter must be 1

Template Parameters

tChannels	The number of data channels of the provided frames
tDetermineTemplateMeans	True, to determine the mean intensities for the current frame and the template frame; False, to determine the mean intensities for the current frame only

See also

determineMeans8BitPerChannel().

optimizeAlignmentAdditive()

static bool Ocean::Tracking::HomographyImageAlignmentDense::optimizeAlignmentAdditive ( const Frame &  templateFrame, const CV::SubRegion &  templateSubRegion, const Frame &  currentFrame, const SquareMatrix3 &  roughHomography, const unsigned int  homographyParameters, const bool  zeroMean, SquareMatrix3 &  homography, const unsigned int  iterations = 20u, Scalar  lambda = 10, const Scalar  lambdaFactor = 10, Scalar *  initialError = nullptr, Scalar *  finalError = nullptr, Scalars *  intermediateErrors = nullptr, ConsistencyDataRef *  externalConsistencyData = nullptr, bool *  abort = nullptr  )

static

Optimizes the alignment between two images within a specified sub-region regarding a homography by application of an iterative additive delta Levenberg-Marquardt optimization approach.

The resulting homography converts points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint), the given rough homography must be defined accordingly.
The performance of this function can be improved further if this function is invoked successively for the same template frame with identical parameters but with individual current frames.
In this case, an empty consistency data object must be provided for the first call, stored outside and must again be provided for the successive calls.

Parameters

templateFrame	The template frame defining the reference to which the current frame will be aligned, must be valid
templateSubRegion	The sub-region defined within the template frame from which the visual content is used for the alignment, image content outside the sub-region is not investigated, must be valid
currentFrame	The current frame for which the alignment will be determine, must have the same pixel format and pixel origin as the template frame
roughHomography	An already rough homography defining the alignment between template and current frame, the homography transforms points defined in the template frame to points defined in the current frame
homographyParameters	The number of parameters used to define the homography, with range [8, 9]
zeroMean	True, to subtract the color intensities from the corresponding mean intensity before determining the error; False, to determine the error directly
homography	The resulting optimized homography for the specified template frame and current frame, with 1 in the lower right corner
iterations	The number of optimization iterations that will be applied at most, with range [1, infinity)
lambda	Initial Levenberg-Marquardt damping value which may be changed after each iteration using the damping factor, with range [0, infinity)
lambdaFactor	Levenberg-Marquardt damping factor to be applied to the damping value, with range [1, infinity)
initialError	Optional resulting initial averaged squared error between template and current frame within the sub-region
finalError	Optional resulting final averaged squared error between template and current frame within the sub-region
intermediateErrors	Optional resulting intermediate averaged squared errors, on error for each improving optimization iteration
abort	Optional abort statement allowing to abort the alignment at any time; set the value True to abort the alignment
externalConsistencyData	Optional abstract consistency data object that may be used to improve the overall performance for several successive optimization calls for individual current frames but with the same template frame

See also

optimizeAlignmentInverseCompositional(), optimizeAlignmentMultiResolution().

optimizeAlignmentInverseCompositional()

static bool Ocean::Tracking::HomographyImageAlignmentDense::optimizeAlignmentInverseCompositional ( const Frame &  templateFrame, const CV::SubRegion &  templateSubRegion, const Frame &  currentFrame, const SquareMatrix3 &  roughHomography, const unsigned int  homographyParameters, const bool  zeroMean, SquareMatrix3 &  homography, const unsigned int  iterations = 20u, Scalar  lambda = 10, const Scalar  lambdaFactor = 10, Scalar *  initialError = nullptr, Scalar *  finalError = nullptr, Scalars *  intermediateErrors = nullptr, ConsistencyDataRef *  externalConsistencyData = nullptr, bool *  abort = nullptr  )

static

Optimizes the alignment between two images within a specified sub-region regarding a homography by application of an iterative inverse compositional delta Levenberg-Marquardt optimization approach.

The resulting homography converts points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint), the given rough homography must be defined accordingly.
The performance of this function can be improved further if this function is invoked successively for the same template frame with identical parameters but with individual current frames.
In this case, an empty consistency data object must be provided for the first call, stored outside and must again be provided for the successive calls.

Parameters

templateFrame	The template frame defining the reference to which the current frame will be aligned, must be valid
templateSubRegion	The sub-region defined within the template frame from which the visual content is used for the alignment, image content outside the sub-region is not investigated, must be valid
currentFrame	The current frame for which the alignment will be determine, must have the same pixel format and pixel origin as the template frame
roughHomography	An already rough homography defining the alignment between template and current frame, the homography transforms points defined in the template frame to points defined in the current frame
homographyParameters	The number of parameters used to define the homography, with range [8, 9]
zeroMean	True, to subtract the color intensities from the corresponding mean intensity before determining the error; False, to determine the error directly
homography	The resulting optimized homography for the specified template frame and current frame, with 1 in the lower right corner
iterations	The number of optimization iterations that will be applied at most, with range [1, infinity)
lambda	Initial Levenberg-Marquardt damping value which may be changed after each iteration using the damping factor, with range [0, infinity)
lambdaFactor	Levenberg-Marquardt damping factor to be applied to the damping value, with range [1, infinity)
initialError	Optional resulting initial averaged squared error between template and current frame within the sub-region
finalError	Optional resulting final averaged squared error between template and current frame within the sub-region
intermediateErrors	Optional resulting intermediate averaged squared errors, on error for each improving optimization iteration
abort	Optional abort statement allowing to abort the alignment at any time; set the value True to abort the alignment
externalConsistencyData	Optional abstract consistency data object that may be used to improve the overall performance for several successive optimization calls for individual current frames but with the same template frame

See also

optimizeAlignmentAdditive(), optimizeAlignmentMultiResolution().

optimizeAlignmentMultiResolution() [1/2]

static bool Ocean::Tracking::HomographyImageAlignmentDense::optimizeAlignmentMultiResolution ( const CV::FramePyramid &  templateFramePyramid, const std::vector< CV::SubRegion > &  templateSubRegions, const CV::FramePyramid &  currentFramePyramid, const unsigned int  layers, const unsigned int  homographyParameters, const bool  additiveAlignment, const bool  levenbergMarquardtOptimization, const bool  zeroMean, const SquareMatrix3 &  roughHomography, SquareMatrix3 &  homography, const unsigned int  coarseIterations = 20u, const unsigned int  fineIterations = 4u, std::vector< Tracking::HomographyImageAlignmentDense::ConsistencyDataRef > *  consistencyDatas = nullptr  )

static

Optimizes the alignment between two images within a specified sub-region regarding a homography by applying a multi-resolution (coarse to fine approach) Gauss-Newton or a Levenberg-Marquardt optimization approach.

The resulting homography converts points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint), the given rough homography must be defined accordingly.
The performance of this function can be improved further if this function is invoked successively for the same template frame with identical parameters but with individual current frames.
In this case, an empty consistency data object must be provided for the first call, stored outside and must again be provided for the successive calls.

Parameters

templateFramePyramid	The pyramid frame of the template frame
templateSubRegions	The multi-resolution sub-regions defined in the template frames (one sub-region for each pyramid-layer)
currentFramePyramid	The pyramid frame of the current frame, must have the same pixel format, pixel origin and layer number as the pyramid of the template frame
layers	The number of pyramid layers to be used for the alignment, with range [1, min(templateFramePyramid.layers(), currentFramePyramid.layers()]
homographyParameters	The number of parameters representing the homography with range [8, 9]
additiveAlignment	True, to apply the additive delta optimization; False, to apply the inverse compositional delta optimization
levenbergMarquardtOptimization	True, to apply the Levenberg-Marquardt optimization; False, to apply Gauss-Newton
zeroMean	True, to apply a zero-mean optimization (color intensities will be subtracted from the mean color intensity); False, to use the color intensities directly
roughHomography	The rough homography already known between the template frame and the current frame (currentPoint = H * templatePoint)
homography	The resulting precise homography
coarseIterations	The number of optimization iterations that will be applied at most on the coarsest pyramid layer, with range [1, infinity)
fineIterations	The number of optimization iterations that will be applied at most on the finest pyramid layer, the layers between coarsest and finest layer will be handled by a linear interpolation between both iteration values, with range [1, infinity)
consistencyDatas	The optional consistency data objects to improve the optimization performance, either no objects or one object per pyramid layer

Returns

True, if succeeded

See also

optimizeAlignmentAdditive(), optimizeAlignmentInverseCompositional().

optimizeAlignmentMultiResolution() [2/2]

static bool Ocean::Tracking::HomographyImageAlignmentDense::optimizeAlignmentMultiResolution ( const Frame &  templateFrame, const CV::SubRegion &  templateSubRegion, const Frame &  currentFrame, const unsigned int  numberPyramidLayers, const unsigned int  homographyParameters, const bool  additiveAlignment, const bool  levenbergMarquardtOptimization, const bool  zeroMean, const SquareMatrix3 &  roughHomography, SquareMatrix3 &  homography, const unsigned int  coarseIterations = 20u, const unsigned int  fineIterations = 4u, const CV::FramePyramid::DownsamplingMode  downsamplingMode = CV::FramePyramid::DM_FILTER_14641  )

static

Optimizes the alignment between two images within a specified sub-region regarding a homography by applying a multi-resolution (coarse to fine approach) Gauss-Newton or a Levenberg-Marquardt optimization approach.

The resulting homography converts points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint), the given rough homography must be defined accordingly.
The performance of this function can be improved further if this function is invoked successively for the same template frame with identical parameters but with individual current frames.
In this case, an empty consistency data object must be provided for the first call, stored outside and must again be provided for the successive calls.

Parameters

templateFrame	The template frame, must be valid
templateSubRegion	The sub-region defined in the template frames, must be valid
currentFrame	The current frame, must have the same pixel format, pixel origin and layer number as the pyramid of the template frame, must be valid
numberPyramidLayers	The number of pyramid layers to be used, with range [1, infinity)
homographyParameters	The number of parameters representing the homography with range [8, 9]
additiveAlignment	True, to apply the additive delta optimization; False, to apply the inverse compositional delta optimization
levenbergMarquardtOptimization	True, to apply the Levenberg-Marquardt optimization; False, to apply Gauss-Newton
zeroMean	True, to apply a zero-mean optimization (color intensities will be subtracted from the mean color intensity); False, to use the color intensities directly
roughHomography	The rough homography already known between the template frame and the current frame (currentPoint = H * templatePoint)
homography	The resulting precise homography
coarseIterations	The number of optimization iterations that will be applied at most on the coarsest pyramid layer, with range [1, infinity)
fineIterations	The number of optimization iterations that will be applied at most on the finest pyramid layer, the layers between coarsest and finest layer will be handled by a linear interpolation between both iteration values, with range [1, infinity)
downsamplingMode	The down-sampling mode that is applied to create the pyramid layers

Returns

True, if succeeded

See also

optimizeAlignmentAdditive(), optimizeAlignmentInverseCompositional().

slowDetermineError()

static Scalar Ocean::Tracking::HomographyImageAlignmentDense::slowDetermineError ( const Frame &  templateFrame, const CV::SubRegion &  templateSubRegion, const Frame &  currentFrame, const SquareMatrix3 &  homography, const bool  zeroMean  )

staticprotected

Determines the current error for a given homography between a current frame and a template frame within a sub-region of the template frame.

The homography transformed points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint).
We keep this implementation for debugging and for better understanding of the overall algorithm.

Parameters

templateFrame	The template frame for which the error is determined, with 8 bit per pixel and 1-4 channels
templateSubRegion	The sub-region defined within the template frame
currentFrame	The current frame for which the error is determined, must have the same pixel format and pixel origin as the template frame
homography	The homography for which the error is determined, transforming points defined in the template frame to points defined in the current frame, with 1 in the lower right corner
zeroMean	True, to subtract the color intensities from the corresponding mean intensity before determining the error; False, to determine the error directly

Returns

The resulting average squared error for all valid pixel correspondences

slowDetermineError8BitPerChannel()

template<unsigned int tChannels>

static Scalar Ocean::Tracking::HomographyImageAlignmentDense::slowDetermineError8BitPerChannel ( const Frame &  templateFrame, const CV::SubRegion &  templateSubRegion, const Frame &  currentFrame, const SquareMatrix3 &  homography, const bool  zeroMean  )

staticprotected

Determines the current error for a given homography between a current frame and a template frame within a sub-region of the template frame.

The homography transformed points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint).
We keep this implementation for debugging and for better understanding of the overall algorithm.

Parameters

templateFrame	The template frame for which the error is determined, with 8 bit per pixel and 1-4 channels
templateSubRegion	The sub-region defined within the template frame
currentFrame	The current frame for which the error is determined, must have the same pixel format and pixel origin as the template frame
homography	The homography for which the error is determined, transforming points defined in the template frame to points defined in the current frame, with 1 in the lower right corner
zeroMean	True, to subtract the color intensities from the corresponding mean intensity before determining the error; False, to determine the error directly

Returns

The resulting average squared error for all valid pixel correspondences

Template Parameters

tChannels

The number of channels both frames have, with range [1, 4]

slowDetermineHessianAndErrorJacobian()

template<unsigned int tParameters>

static bool Ocean::Tracking::HomographyImageAlignmentDense::slowDetermineHessianAndErrorJacobian ( const Frame &  templateFrame, const CV::SubRegion &  templateSubRegion, const Frame &  currentFrame, const SquareMatrix3 &  homography, const bool  zeroMean, Matrix &  hessian, Matrix &  jacobianError  )

staticprotected

Determines the 8x8 or 9x9 Hessian matrix and the 8x1 or 9x1 Jacobian-Error vector for a given homography between a current frame and a template frame within a sub-region of the template frame.

The homography transformed points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint).
We keep this implementation for debugging and for better understanding of the overall algorithm.

Parameters

templateFrame	The template frame for which the Hessian and Jacobian is determined, with 8 bit per pixel and 1-4 channels
templateSubRegion	The sub-region defined within the template frame
currentFrame	The current frame for which the Hessian and Jacobian is determined, must have the same pixel format and pixel origin as the template frame
homography	The homography for which the Hessian and Jacobian is determined, transforming points defined in the template frame to points defined in the current frame, with 1 in the lower right corner
zeroMean	True, to subtract the color intensities from the corresponding mean intensity before determining the error; False, to determine the error directly
hessian	The resulting Hessian matrix
jacobianError	The resulting Jacobian-Error matrix

Returns

True, if succeeded

Template Parameters

tParameters

The number of parameters describing the homography and used to optimize it, with range [8, 9]

slowDetermineHessianAndErrorJacobian8BitPerChannel()

template<unsigned int tParameters, unsigned int tChannels>

static bool Ocean::Tracking::HomographyImageAlignmentDense::slowDetermineHessianAndErrorJacobian8BitPerChannel ( const Frame &  templateFrame, const CV::SubRegion &  templateSubRegion, const Frame &  currentFrame, const SquareMatrix3 &  homography, const bool  zeroMean, Matrix &  hessian, Matrix &  jacobianError  )

staticprotected

Determines the 8x8 or 9x9 Hessian matrix and the 8x1 or 9x1 Jacobian-Error vector for a given homography between a current frame and a template frame within a sub-region of the template frame.

The homography transformed points defined in the template frame to points defined in the current frame (currenttPoint = H * templatePoint).
We keep this implementation for debugging and for better understanding of the overall algorithm.

Parameters

templateFrame	The template frame for which the Hessian and Jacobian is determined, with 8 bit per pixel and 1-4 channels
templateSubRegion	The sub-region defined within the template frame
currentFrame	The current frame for which the Hessian and Jacobian is determined, must have the same pixel format and pixel origin as the template frame
homography	The homography for which the Hessian and Jacobian is determined, transforming points defined in the template frame to points defined in the current frame, with 1 in the lower right corner
hessian	The resulting Hessian matrix
zeroMean	True, to subtract the color intensities from the corresponding mean intensity before determining the error; False, to determine the error directly
jacobianError	The resulting Jacobian-Error matrix

Returns

True, if succeeded

Template Parameters

tParameters	The number of parameters describing the homography and used to optimize it, with range [8, 9]
tChannels	The number of channels both frames have, with range [1, 4]

---

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

• HomographyImageAlignmentDense.h