#include <vpRobust.h>

Public Types
enum	vpRobustEstimatorType { TUKEY, CAUCHY, HUBER }

Public Member Functions
	vpRobust (unsigned int n_data)

virtual	~vpRobust ()

void	resize (unsigned int n_data)

void	MEstimator (const vpRobustEstimatorType method, const vpColVector &residues, vpColVector &weights)

void	MEstimator (const vpRobustEstimatorType method, const vpColVector &residues, const vpColVector &all_residues, vpColVector &weights)

vpColVector	simultMEstimator (vpColVector &residues)

void	setIteration (const unsigned int iter)

void	setThreshold (const double noise_threshold)

Detailed Description

Contains an M-Estimator and various influence function.

Supported methods: M-estimation, Tukey, Cauchy and Huber

Definition at line 63 of file vpRobust.h.

Member Enumeration Documentation

Enumeration of influence functions.

Definition at line 68 of file vpRobust.h.

vpRobust::vpRobust ( unsigned int n_data )

Default Constructor.

Constructor.

Parameters

n_data : Size of the data vector.

Definition at line 72 of file vpRobust.cpp.

virtual vpRobust::~vpRobust ( )

inlinevirtual

Destructor.

Definition at line 102 of file vpRobust.h.

void vpRobust::MEstimator	(	const vpRobustEstimatorType	method,
		const vpColVector &	residues,
		vpColVector &	weights
	)

Compute the weights according a residue vector and a PsiFunction.

Calculate an Mestimate given a particular loss function using MAD (Median Absolute Deviation) as a scale estimate at each iteration.

Postcondition: Keeps a copy of the weights so that rejected points are kept at zero weight.

Parameters

method : Type of M-Estimator $\rho(r_i)$ :

TUKEY : $\rho(r_i, C) = \left\{ \begin{array}{ll} \frac{r_i^6}{6} - \frac{C^2r_i^4}{2} +\frac{C^4r_i^2}{2} & \mbox{if} |r_i| < C \\ \frac{1}{6} C^6 & \mbox{else} \end{array} \right.$ with influence function $\psi(r_i, C) = \left\{ \begin{array}{ll} r_i(r_i^2-C^2)^2 & \mbox{if} |r_i| < C \\ 0 & \mbox{else} \end{array} \right.$ where $C=4.7 \hat{\sigma}$ and with $\hat{\sigma} = 1.48{Med}_i(|r_i - {Med}_j(r_j)|)$
CAUCHY :
HUBER :

Parameters

residues	: Residues used in the previous formula.
weights	: Vector of weights $w_i = \frac{\psi(r_i)}{r_i}$ . Values are in [0, 1]. A value near zero means that the data is an outlier. This vector must have the same size residue vector.