Non-linear optimization routines. There are three methods implemented that include Gauss-Newton, Levenberg-Marquardt and Tukey m-estimator. More...

Public Types
typedef void(*	EstimateCallback )(CvMat state, CvMat projection, void *param)
	Pointer to the function that projects the state of the system to the measurements.
enum	OptimizeMethod { GAUSSNEWTON, LEVENBERGMARQUARDT, TUKEY_LM }
	Selection between the algorithm used in optimization. Following should be noticed: More...
Public Member Functions
void	CalcJacobian (CvMat x, CvMat J, EstimateCallback Estimate)
	Numerically differentiates and calculates the Jacobian around x.
CvMat *	GetErr ()
	Returns the current residual vector.
	Optimization (int n_params, int n_meas)
	Constructor.
double	Optimize (CvMat parameters, CvMat measurements, double stop, int max_iter, EstimateCallback Estimate, void param=0, OptimizeMethod method=LEVENBERGMARQUARDT, CvMat parameters_mask=0, CvMat J_mat=0, CvMat weights=0)
	Runs the optimization loop with selected parameters.
	~Optimization ()
Private Member Functions
double	CalcTukeyWeight (double residual, double c)
double	CalcTukeyWeightSimple (double residual, double c)
Private Attributes
CvMat *	delta
CvMat *	diag
CvMat *	err
void *	estimate_param
CvMat *	J
CvMat *	JtJ
double	lambda
CvMat *	tmp
CvMat *	tmp_par
CvMat *	W
CvMat *	x_minus
CvMat *	x_plus
CvMat *	x_tmp1
CvMat *	x_tmp2

Detailed Description

Non-linear optimization routines. There are three methods implemented that include Gauss-Newton, Levenberg-Marquardt and Tukey m-estimator.

Definition at line 44 of file Optimization.h.

Member Typedef Documentation

typedef void(* alvar::Optimization::EstimateCallback)(CvMat *state, CvMat *projection, void *param)

Pointer to the function that projects the state of the system to the measurements.

Parameters:

state	System parameters, e.g. camera parameterization in optical tracking.
projection	The system state projection is stored here. E.g image measurements in optical tracking.
param	Additional parameters to the function. E.g. some constant parameters that are not optimized.

Definition at line 101 of file Optimization.h.

Selection between the algorithm used in optimization. Following should be noticed:

Enumerator:

Definition at line 74 of file Optimization.h.

alvar::Optimization::Optimization	(	int	n_params,
		int	n_meas
	)

Constructor.

Parameters:

n_params	Number of parameters to be optimized.
n_meas	Number of measurements that are observed.

Definition at line 34 of file Optimization.cpp.

alvar::Optimization::~Optimization ( )

Definition at line 51 of file Optimization.cpp.

void alvar::Optimization::CalcJacobian	(	CvMat *	x,
		CvMat *	J,
		EstimateCallback	Estimate
	)

Numerically differentiates and calculates the Jacobian around x.

Parameters:

x	The set of parameters around which the Jacobian is evaluated.
J	Resulting Jacobian matrix is stored here.
Estimate	The function to be differentiated.

Definition at line 99 of file Optimization.cpp.

double alvar::Optimization::CalcTukeyWeight	(	double	residual,
		double	c
	)		`[private]`

Definition at line 68 of file Optimization.cpp.

double alvar::Optimization::CalcTukeyWeightSimple	(	double	residual,
		double	c
	)		`[private]`

Definition at line 89 of file Optimization.cpp.

CvMat* alvar::Optimization::GetErr ( ) [inline]

Returns the current residual vector.

Definition at line 93 of file Optimization.h.

double alvar::Optimization::Optimize	(	CvMat *	parameters,
		CvMat *	measurements,
		double	stop,
		int	max_iter,
		EstimateCallback	Estimate,
		void *	param = `0`,
		OptimizeMethod	method = `LEVENBERGMARQUARDT`,
		CvMat *	parameters_mask = `0`,
		CvMat *	J_mat = `0`,
		CvMat *	weights = `0`
	)

Runs the optimization loop with selected parameters.

Parameters:

parameters	Vector of parameters to be optimized. Initial values should be set.
measurements	Vector of measurements that are observed.
stop	Optimization loop ends as the stop limit is reached. Criteria is calculated as
max_iter	Maximum number of iteration loops that are evaluated if stop is not reached.
Estimate	Pointer to the function that maps the state to the measurements. See EstimateCallback.
method	One of the three possible optimization methods.
parameters_mask	Vector that defines the parameters that are optimized. If vector element is 0, corresponding parameter is not altered.
J_mat	Jacobian matrix. If not given, numerical differentation is used.
weights	Weight vector that can be submitted to give different weights to different measurements. Currently works only with OptimizeMethod::TUKEY_LM.