opentl::core::util::GaussNewton Class Reference

Gauss-Newton optimization. This class implements a single Gauss-Newton optimization step for nonlinear least-squares problems. For efficiency reasons, it uses cached data structures, related to the measurement size, and the number of pose degrees of freedom. More...

Inherits opentl::core::util::ParameterContainer.

List of all members.

Classes
struct	GaussNewtonData_Key_dof
struct	GaussNewtonData_Key_dof_and_featSize
	structure for caching individual GN-terms (single features). key = pose dof (left 16bits) AND feature size (right 16bits) More...
Public Types
Public Member Functions
	GaussNewton ()
	Constructor.
double	gaussNewtonUpdate (const std::vector< core::cvdata::T_MEAS_FEAT * > &measurements, opentl::core::cvdata::Pose &currPose, opentl::math::Vector *priorGrad=NULL, opentl::math::SquareMatrix *priorHess=NULL, opentl::math::SquareMatrix *estCov=NULL)
	Perform Gauss-Newton parameter update (used in order to upgrade from feature to object level) for all targets.
void	init ()
	Initialization.
	~GaussNewton ()

---

Detailed Description

Gauss-Newton optimization. This class implements a single Gauss-Newton optimization step for nonlinear least-squares problems. For efficiency reasons, it uses cached data structures, related to the measurement size, and the number of pose degrees of freedom.

---

Member Typedef Documentation

typedef struct opentl::core::util::GaussNewton::GaussNewtonData_Key_dof opentl::core::util::GaussNewton::T_GAUSSNEWTON_DATA_KEY_DOF

typedef struct opentl::core::util::GaussNewton::GaussNewtonData_Key_dof_and_featSize opentl::core::util::GaussNewton::T_GAUSSNEWTON_DATA_KEY_DOF_AND_FEATSIZE

structure for caching individual GN-terms (single features). key = pose dof (left 16bits) AND feature size (right 16bits)

---

Member Enumeration Documentation

enum opentl::core::util::GaussNewton::OfflineParameters

Enumerator:

OFFLINE_COUNT

enum opentl::core::util::GaussNewton::OnlineParameters

Enumerator:

param_MEstimator	M-estimator for robust LSE.
param_MEstimatorThresh	M-estimator parameter (threshold for outliers).
param_enableDampingFactor	flag to enable damping factor in case of a singular matrix
param_dampingFactorDeterminantMinThreshold	min. threshold of Hessian determinant value for enabling damping
param_dampingFactorMax	maximum damping factor
ONLINE_COUNT

enum opentl::core::util::GaussNewton::T_GAUSSNEWTON_MEST

Possible M-estimators.

Enumerator:

NONE	Standard LSE.
TUKEY	Tukey estimator.
HUBER	Huber estimator.
EDGE_SALIENCY	Edge saliency correction (see the paper of Drummond).

---

Constructor & Destructor Documentation

opentl::core::util::GaussNewton::GaussNewton

(

)

Constructor.

opentl::core::util::GaussNewton::~GaussNewton

(

)

---

Member Function Documentation

double opentl::core::util::GaussNewton::gaussNewtonUpdate	(	const std::vector< core::cvdata::T_MEAS_FEAT * > &	measurements,
		opentl::core::cvdata::Pose &	currPose,
		opentl::math::Vector *	priorGrad = NULL,
		opentl::math::SquareMatrix *	priorHess = NULL,
		opentl::math::SquareMatrix *	estCov = NULL
	)

Perform Gauss-Newton parameter update (used in order to upgrade from feature to object level) for all targets.

Parameters:

	measurements	Input vector of multi-modal measurements and Jacobians (feature-level) related to a single target
	currPose	Input and output pose (updated with Gauss-Newton in the incremental parameters)
	priorGrad	Input: gradient of the prior distribution
	priorHess	Input: Hessian of the prior distribution
	estCov	Output: covariance of the estimate (= invert Gauss-Newton matrix)

Returns:

Norm of incremental vector

void opentl::core::util::GaussNewton::init

(

)

[inline, virtual]

Initialization.

Implements opentl::core::util::ParameterContainer.