opentl::tracker::Tracker Class Reference

Abstract base class for all trackers to come. More...

Inherited by opentl::tracker::DummyTracker, opentl::tracker::InfoFilter, opentl::tracker::Kalman, opentl::tracker::MCMCParticle, opentl::tracker::SIRParticle, and opentl::tracker::UnscentedInfoFilter.

List of all members.

Public Types

Public Member Functions

virtual void addExplicitLikelihoodInstances (std::size_t count=1)

Add additional likelihood instances (for multiple threads) Instances ares cloned from the likelihood template and represent the whole pipeline.

virtual void addImplicitLikelihoodInstances (std::size_t count=1)

virtual void computeOutputState (opentl::models::Target &target, std::size_t burnInSamples=0)

Compute output state (weighted particle mean).

virtual void computeOutputStateAndPoseCovariance (opentl::models::Target &target, math::SquareMatrix *poseCov=NULL, std::size_t burnInSamples=0)

Compute weighted particle mean and covariance (of the finite pose parameters only).

virtual void correct (TargetPtrVector &targets)=0

Correct: perform a measurement (using the Likelihood) and update the posterior distribution of all targets.

virtual bool detectLoss (boost::shared_ptr< opentl::models::Target > target, double value, TrackLoss method=DETERMINANT)

Function to detect a target loss.

core::cvdata::T_MEAS_LIK * getExplicitMeas (std::size_t threadNum, std::size_t targetNum)

Obtain the explicit measurement for a given thread and target.

opentl::modalities::Likelihood::T_LIK_MODALITY_FEATURE * getInternalFeatModality (int m, std::size_t threadId) const

opentl::modalities::Likelihood::T_LIK_MODALITY_OBJECT * getInternalObjModality (int m, std::size_t threadId) const

opentl::modalities::Likelihood::T_LIK_MODALITY_PIXEL * getInternalPixModality (int m, std::size_t threadId) const

opentl::modalities::Likelihood * getLikelihood (std::size_t i)

virtual void init (TargetPtrVector &targets, std::vector< boost::shared_ptr< opentl::core::State > > *initStates=NULL, std::vector< boost::shared_ptr< math::SquareMatrix > > *initCovs=NULL, std::vector< int > *nHypotheses=NULL)=0

Initialize: allocate internal data, and update warp, for all targets.

virtual void predict (TargetPtrVector &targets)=0

Predict: compute the prior distribution of all targets from the last posterior, using the dynamical model.

virtual void resizeExplicitMeasurements (int nOfTargets)

resize the measurements for each likelihood instance (for multiple targets

int sampleModelFeatures (const TargetPtrVector &targets, std::size_t threadId)

Call (recursively) the sampleModelFeatures method for the whole processing tree.

Tracker (const opentl::modalities::Likelihood &likelihood, opentl::modelprojection::Warp &warp)

constructor (objectModels are created dynamically) Use this constructor for varying targets

int updateModelFeatures (const TargetPtrVector &targets, std::size_t threadId)

Call (recursively) the updateModelFeatures method for the whole processing tree.

virtual ~Tracker ()

destructor

Protected Member Functions

virtual void output (TargetPtrVector &targets)

Compute the output state for all targets.

Protected Attributes

std::vector< boost::shared_ptr
< modalities::Likelihood > > mLikelihoods

const
opentl::modalities::Likelihood & mLikelihoodTemplate

std::vector< T_MEAS_LIKPtrVector > mLikMeasExplicit

opentl::modelprojection::Warp & mWarp

Main warp class, needed because the correction and prediction steps call the warpUpdate() for the respective state vectors.

Detailed Description

Abstract base class for all trackers to come.

Author:

The aim of this class is, to provide the two main tracking functions PREDICT and CORRECT All inherited trackers should implement these functions, so the user (or the algorithm) does not need to know how the filter works exactly but has just to call these functions.

Member Enumeration Documentation

enum opentl::tracker::Tracker::TrackLoss

Method to determine a lost track.

Enumerator:

DETERMINANT	Use the determinant of the covariance matrix.
LIKELIHOOD	Use the internal weights (likelihoods) - only for particle filters.

Constructor & Destructor Documentation

opentl::tracker::Tracker::Tracker	(	const opentl::modalities::Likelihood &	likelihood,
		opentl::modelprojection::Warp &	warp
	)

constructor (objectModels are created dynamically) Use this constructor for varying targets

virtual opentl::tracker::Tracker::~Tracker ( ) [virtual]

destructor

Member Function Documentation

virtual void opentl::tracker::Tracker::addExplicitLikelihoodInstances ( std::size_t count = 1 ) [virtual]

Add additional likelihood instances (for multiple threads) Instances ares cloned from the likelihood template and represent the whole pipeline.

virtual void opentl::tracker::Tracker::addImplicitLikelihoodInstances ( std::size_t count = 1 ) [virtual]

virtual void opentl::tracker::Tracker::computeOutputState	(	opentl::models::Target &	target,
		std::size_t	burnInSamples = `0`
	)			`[virtual]`

Compute output state (weighted particle mean).

virtual void opentl::tracker::Tracker::computeOutputStateAndPoseCovariance	(	opentl::models::Target &	target,
		math::SquareMatrix *	poseCov = `NULL`,
		std::size_t	burnInSamples = `0`
	)			`[virtual]`

Compute weighted particle mean and covariance (of the finite pose parameters only).

Parameters:

	target	Single target distribution
	poseCov	(optional) Output pose covariance
	burnInSamples	(optional) Burn-in samples, to be discarded from the computation

virtual void opentl::tracker::Tracker::correct ( TargetPtrVector & targets ) [pure virtual]

Correct: perform a measurement (using the Likelihood) and update the posterior distribution of all targets.

Parameters:

targets

Vector of targets

Implemented in opentl::tracker::DummyTracker, opentl::tracker::InfoFilter, opentl::tracker::Kalman, opentl::tracker::MCMCParticle, opentl::tracker::SIRParticle, and opentl::tracker::UnscentedInfoFilter.

virtual bool opentl::tracker::Tracker::detectLoss	(	boost::shared_ptr< opentl::models::Target >	target,
		double	value,
		TrackLoss	method = `DETERMINANT`
	)			`[virtual]`

Function to detect a target loss.

core::cvdata::T_MEAS_LIK* opentl::tracker::Tracker::getExplicitMeas	(	std::size_t	threadNum,
		std::size_t	targetNum
	)

Obtain the explicit measurement for a given thread and target.

Parameters:

	threadNum	Index of the target
	targetNum	Index of the target

Returns:: Pointer to the explicit measurement structure

opentl::modalities::Likelihood::T_LIK_MODALITY_FEATURE* opentl::tracker::Tracker::getInternalFeatModality	(	int	m,
		std::size_t	threadId
	)			const

opentl::modalities::Likelihood::T_LIK_MODALITY_OBJECT* opentl::tracker::Tracker::getInternalObjModality	(	int	m,
		std::size_t	threadId
	)			const

opentl::modalities::Likelihood::T_LIK_MODALITY_PIXEL* opentl::tracker::Tracker::getInternalPixModality	(	int	m,
		std::size_t	threadId
	)			const

opentl::modalities::Likelihood* opentl::tracker::Tracker::getLikelihood ( std::size_t i )

virtual void opentl::tracker::Tracker::init	(	TargetPtrVector &	targets,
		std::vector< boost::shared_ptr< opentl::core::State > > *	initStates = `NULL`,
		std::vector< boost::shared_ptr< math::SquareMatrix > > *	initCovs = `NULL`,
		std::vector< int > *	nHypotheses = `NULL`
	)			`[pure virtual]`

Initialize: allocate internal data, and update warp, for all targets.

Parameters:

	targets	Vector of targets
	initStates	Vector of initial states (optional); it must have same size of targets
	initCovs	Vector of initial covariances (optional); it must have the same size of targets
	nHypotheses	Vector with number of state hypotheses for each target distribution; it must have same size of targets

Implemented in opentl::tracker::DummyTracker, opentl::tracker::InfoFilter, opentl::tracker::Kalman, opentl::tracker::MCMCParticle, opentl::tracker::SIRParticle, and opentl::tracker::UnscentedInfoFilter.

virtual void opentl::tracker::Tracker::output ( TargetPtrVector & targets ) [protected, virtual]

Compute the output state for all targets.

Parameters:

targets

Vector of targets

Reimplemented in opentl::tracker::DummyTracker, opentl::tracker::InfoFilter, opentl::tracker::Kalman, opentl::tracker::MCMCParticle, opentl::tracker::SIRParticle, and opentl::tracker::UnscentedInfoFilter.

virtual void opentl::tracker::Tracker::predict ( TargetPtrVector & targets ) [pure virtual]

Predict: compute the prior distribution of all targets from the last posterior, using the dynamical model.

Parameters:

targets

Vector of targets

Implemented in opentl::tracker::DummyTracker, opentl::tracker::InfoFilter, opentl::tracker::Kalman, opentl::tracker::MCMCParticle, opentl::tracker::SIRParticle, and opentl::tracker::UnscentedInfoFilter.

virtual void opentl::tracker::Tracker::resizeExplicitMeasurements ( int nOfTargets ) [virtual]

resize the measurements for each likelihood instance (for multiple targets

int opentl::tracker::Tracker::sampleModelFeatures	(	const TargetPtrVector &	targets,
		std::size_t	threadId
	)

Call (recursively) the sampleModelFeatures method for the whole processing tree.

int opentl::tracker::Tracker::updateModelFeatures	(	const TargetPtrVector &	targets,
		std::size_t	threadId
	)

Call (recursively) the updateModelFeatures method for the whole processing tree.

Member Data Documentation

std::vector<boost::shared_ptr<modalities::Likelihood> > opentl::tracker::Tracker::mLikelihoods [protected]

const opentl::modalities::Likelihood& opentl::tracker::Tracker::mLikelihoodTemplate [protected]

std::vector<T_MEAS_LIKPtrVector> opentl::tracker::Tracker::mLikMeasExplicit [protected]

opentl::modelprojection::Warp& opentl::tracker::Tracker::mWarp [protected]

Main warp class, needed because the correction and prediction steps call the warpUpdate() for the respective state vectors.


Public Types
Public Member Functions
virtual void	addExplicitLikelihoodInstances (std::size_t count=1)
	Add additional likelihood instances (for multiple threads) Instances ares cloned from the likelihood template and represent the whole pipeline.
virtual void	addImplicitLikelihoodInstances (std::size_t count=1)
virtual void	computeOutputState (opentl::models::Target &target, std::size_t burnInSamples=0)
	Compute output state (weighted particle mean).
virtual void	computeOutputStateAndPoseCovariance (opentl::models::Target &target, math::SquareMatrix *poseCov=NULL, std::size_t burnInSamples=0)
	Compute weighted particle mean and covariance (of the finite pose parameters only).
virtual void	correct (TargetPtrVector &targets)=0
	Correct: perform a measurement (using the Likelihood) and update the posterior distribution of all targets.
virtual bool	detectLoss (boost::shared_ptr< opentl::models::Target > target, double value, TrackLoss method=DETERMINANT)
	Function to detect a target loss.
core::cvdata::T_MEAS_LIK *	getExplicitMeas (std::size_t threadNum, std::size_t targetNum)
	Obtain the explicit measurement for a given thread and target.
opentl::modalities::Likelihood::T_LIK_MODALITY_FEATURE *	getInternalFeatModality (int m, std::size_t threadId) const
opentl::modalities::Likelihood::T_LIK_MODALITY_OBJECT *	getInternalObjModality (int m, std::size_t threadId) const
opentl::modalities::Likelihood::T_LIK_MODALITY_PIXEL *	getInternalPixModality (int m, std::size_t threadId) const
opentl::modalities::Likelihood *	getLikelihood (std::size_t i)
virtual void	init (TargetPtrVector &targets, std::vector< boost::shared_ptr< opentl::core::State > > initStates=NULL, std::vector< boost::shared_ptr< math::SquareMatrix > > initCovs=NULL, std::vector< int > *nHypotheses=NULL)=0
	Initialize: allocate internal data, and update warp, for all targets.
virtual void	predict (TargetPtrVector &targets)=0
	Predict: compute the prior distribution of all targets from the last posterior, using the dynamical model.
virtual void	resizeExplicitMeasurements (int nOfTargets)
	resize the measurements for each likelihood instance (for multiple targets
int	sampleModelFeatures (const TargetPtrVector &targets, std::size_t threadId)
	Call (recursively) the sampleModelFeatures method for the whole processing tree.
	Tracker (const opentl::modalities::Likelihood &likelihood, opentl::modelprojection::Warp &warp)
	constructor (objectModels are created dynamically) Use this constructor for varying targets
int	updateModelFeatures (const TargetPtrVector &targets, std::size_t threadId)
	Call (recursively) the updateModelFeatures method for the whole processing tree.
virtual	~Tracker ()
	destructor
Protected Member Functions
virtual void	output (TargetPtrVector &targets)
	Compute the output state for all targets.
Protected Attributes
std::vector< boost::shared_ptr < modalities::Likelihood > >	mLikelihoods
const opentl::modalities::Likelihood &	mLikelihoodTemplate
std::vector< T_MEAS_LIKPtrVector >	mLikMeasExplicit
opentl::modelprojection::Warp &	mWarp
	Main warp class, needed because the correction and prediction steps call the warpUpdate() for the respective state vectors.