osvr::kalman::SigmaPointCorrectionApplication< State, Measurement > Class Template Reference

Public Types
using	StateVec = types::DimVector< State >
using	StateSquareMatrix = types::DimSquareMatrix< State >
using	MeasurementVec = types::DimVector< Measurement >
using	MeasurementSquareMatrix = types::DimSquareMatrix< Measurement >
using	AugmentedStateVec = types::Vector< AugmentedStateDim >
using	AugmentedStateCovMatrix = types::SquareMatrix< AugmentedStateDim >
using	SigmaPointsGen = AugmentedSigmaPointGenerator< AugmentedStateDim, n >
using	Reconstruction = ReconstructedDistributionFromSigmaPoints< m, SigmaPointsGen >
using	TransformedSigmaPointsMat = typename Reconstruction::TransformedSigmaPointsMat
using	GainMatrix = types::Matrix< n, m >

Public Member Functions
	SigmaPointCorrectionApplication (State &s, Measurement &m, SigmaPointParameters const &params=SigmaPointParameters())
bool	finishCorrection (bool cancelIfNotFinite=true)

Static Public Member Functions
static AugmentedStateVec	getAugmentedStateVec (State const &s, Measurement const &m)
static AugmentedStateCovMatrix	getAugmentedStateCov (State const &s, Measurement &meas)
static TransformedSigmaPointsMat	transformSigmaPoints (State const &s, Measurement &meas, SigmaPointsGen const &sigmaPoints)
static MeasurementSquareMatrix	computeInnovationCovariance (State const &s, Measurement &meas, Reconstruction const &recon)
static StateVec	computeStateCorrection (Reconstruction const &recon, MeasurementVec const &deltaz, Eigen::LDLT< MeasurementSquareMatrix > const &pvvDecomp)

Public Attributes
State &	state
Measurement &	measurement
SigmaPointsGen	sigmaPoints
TransformedSigmaPointsMat	transformedPoints
Reconstruction	reconstruction
MeasurementSquareMatrix	innovationCovariance
	aka Pvv
Eigen::LDLT < MeasurementSquareMatrix >	PvvDecomp
types::Vector< m >	deltaz
	reconstructed mean measurement residual/delta z/innovation
StateVec	stateCorrection
bool	stateCorrectionFinite

Static Public Attributes
static const types::DimensionType	n = types::Dimension<State>::value
static const types::DimensionType	m
static const types::DimensionType	AugmentedStateDim = n + m
	state augmented with measurement noise mean
static const types::DimensionType	NumSigmaPoints

Detailed Description

template<typename State, typename Measurement>
class osvr::kalman::SigmaPointCorrectionApplication< State, Measurement >

Definition at line 49 of file FlexibleUnscentedCorrect.h.

Member Function Documentation

template<typename State, typename Measurement>

static AugmentedStateVec osvr::kalman::SigmaPointCorrectionApplication< State, Measurement >::getAugmentedStateVec ( State const &  s, Measurement const &  m  )

inlinestatic

assuming measurement noise is zero mean

Definition at line 108 of file FlexibleUnscentedCorrect.h.

template<typename State, typename Measurement>

static TransformedSigmaPointsMat osvr::kalman::SigmaPointCorrectionApplication< State, Measurement >::transformSigmaPoints ( State const &  s, Measurement &  meas, SigmaPointsGen const &  sigmaPoints  )

inlinestatic

Transforms sigma points by having the measurement class compute the estimated measurement for a state whose state vector we update to each of the sigma points in turn.

Definition at line 128 of file FlexibleUnscentedCorrect.h.

template<typename State, typename Measurement>

bool osvr::kalman::SigmaPointCorrectionApplication< State, Measurement >::finishCorrection ( bool  cancelIfNotFinite = true )

inline

Finish computing the rest and correct the state.

Parameters

cancelIfNotFinite

If the new error covariance is detected to contain non-finite values, should we cancel the correction and not apply it?

Returns

true if correction completed

Logically state.errorCovariance() - K * Pvv * K.transpose(), but considering just the second term, we can replace K with its definition (Pxv Pvv^-1), distribute the transpose on the right over the product, then pull out Pvv^-1 * Pvv * (Pvv^-1).transpose() as "B", leaving Pxv B Pxv.transpose()

Since innovationCovariance aka Pvv is symmetric, (Pvv^-1).transpose() = Pvv^-1. Left multiplication gives Pvv B = Pvv * Pvv^-1 * Pvv * Pvv^-1 whose right hand side is the Pvv-sized identity, and that is in a form that allows us to use our existing LDLT decomp of Pvv to solve for B then evaluate the full original expression.

Definition at line 170 of file FlexibleUnscentedCorrect.h.

Member Data Documentation

template<typename State, typename Measurement>

const types::DimensionType osvr::kalman::SigmaPointCorrectionApplication< State, Measurement >::m

static

Initial value:

=
            types::Dimension<Measurement>::value

Definition at line 58 of file FlexibleUnscentedCorrect.h.

template<typename State, typename Measurement>

const types::DimensionType osvr::kalman::SigmaPointCorrectionApplication< State, Measurement >::NumSigmaPoints

static

Initial value:

=
            SigmaPointsGen::NumSigmaPoints

Definition at line 73 of file FlexibleUnscentedCorrect.h.

---

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

• plugins/unifiedvideoinertialtracker/FlexibleUnscentedCorrect.h