de.unihalle.informatik.MiToBo.tracking.multitarget.distributions.impl

Class AssociationDistributionNN<S extends TargetID,T extends TargetID>

java.lang.Object

de.unihalle.informatik.MiToBo.tracking.multitarget.distributions.abstracts.AbstractAssociationDistribution<S,T>

de.unihalle.informatik.MiToBo.tracking.multitarget.distributions.impl.AssociationDistribution<S,T>

de.unihalle.informatik.MiToBo.tracking.multitarget.distributions.impl.AssociationDistributionNN<S,T>

Type Parameters:

S - Type of discrete variables in the multi target observation

T - Type of discrete variables in the multi target state

All Implemented Interfaces:

EvaluatableDistribution<DataAssociation>, LogEvaluatableDistribution<DataAssociation>, SamplingDistribution<DataAssociation>

public class AssociationDistributionNN<S extends TargetID,T extends TargetID>
extends AssociationDistribution<S,T>

Association distribution to sample association variables for a set of observations based on a model of how observations are formed.

Observations are comprised from existing targets that are detected with probability P_D, a number of observations of newborn targets distributed according to a distribution nu and a number of clutter observations distributed according to a distribution mu.

The association variables of the individual observations are sampled sequentially and their distributions are assumed to depend on the likelihood of the current observation for a specific realization of the association variable and the probability of the association variable given all previous associations. Furthermore the likelihood of nearby observations is considered. The number of the considered neighboring observations can be controlled by a maximum number as well as a maximum Euklidean distance. Keep the combinatorial problem in mind that explodes when the number of considered neighbors increases!!

Author:

Oliver Gress

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	AssociationDistributionNN.AType
private class	AssociationDistributionNN.ObsDistance
private class	AssociationDistributionNN.ProbTreeData

Field Summary

Fields

Modifier and Type	Field and Description
protected AssociationDistributionNN.ObsDistance[][]	kNearestObs Stores for each observation the k-nearest observations that will be associated after the corresponding observation.
protected LogFaculty	logFac An object to compute and store log(n!)
protected double	maxDistNeighbors Maximum distance of neighboring observations
protected int	maxNumNeighbors Maximum number of neighboring observations

Fields inherited from class de.unihalle.informatik.MiToBo.tracking.multitarget.distributions.impl.AssociationDistribution

chi, lastM, lastN, lastSample, logBinom, logMuValues, logNuValues, logP_C, logP_MN, M_max, minMN, mu, newtargetID, nu, P_D, phi_0, phi_1, psi

Fields inherited from class de.unihalle.informatik.MiToBo.tracking.multitarget.distributions.abstracts.AbstractAssociationDistribution

assocfactory, clutterdistrib, log_pzc, M, N, newborndistrib, obsdistrib, rand, Z

Constructor Summary

Constructors

Constructor and Description
AssociationDistributionNN(Random rand, AbstractMultiState<S> Z, AbstractMultiObservationDistributionIndep<S,T> observationDistrib, LogProbabilityDensityFunction spatialClutterDistrib, LogProbabilityDensityFunction spatialNewbornDistrib, LogProbabilityMassFunction mu, LogProbabilityMassFunction nu, double P_D, int maxNumNeighbors, double maxDistNeighbors) Constructor.
AssociationDistributionNN(Random rand, AbstractMultiState<S> Z, AbstractMultiObservationDistributionIndep<S,T> observationDistrib, LogProbabilityDensityFunction spatialClutterDistrib, LogProbabilityDensityFunction spatialNewbornDistrib, LogProbabilityMassFunction mu, LogProbabilityMassFunction nu, double P_D, int M_max, int maxNumNeighbors, double maxDistNeighbors) Constructor where the maximum number of observations in the time series is specified to avoid some re-computations.

Method Summary

Modifier and Type	Method and Description
private double	compute_pczAhead(MTBTreeNode pcTree, Vector<Integer> availableTargets, Stack<Integer> observations) Compute probability of all possible associations of neighboring observations including their likelihood
private MTBTreeNode	compute_qAhead(int m, int mmax, int kmin, int bmin, double logPc_previous, MTBTreeNode subtree) Compute tree of data association prior probabilities depending on previous associations.
DataAssociation	drawSample() Generate a new sample from this density.
DataAssociation	drawSampleDebug(DataAssociation groundtruth, OutputStream ostream)
protected AssociationDistributionNN.ObsDistance[][]	kNearestObservations(int maxNumNeighbors, double maxDistNeighbors) Get the (k) nearest observations for each observation in this.Z, i.e. for each observation z_m all following observations z_{m:M} are sorted by Euklidean distance and stored in an array.
void	setNewObservations(AbstractMultiState<S> Z, AbstractMultiObservationDistributionIndep<S,T> observationDistrib)

Methods inherited from class de.unihalle.informatik.MiToBo.tracking.multitarget.distributions.impl.AssociationDistribution

letNewbornTargetIDsStartFrom, log_p, p, reset

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

kNearestObs

protected AssociationDistributionNN.ObsDistance[][] kNearestObs

Stores for each observation the k-nearest observations that will be associated after the corresponding observation.

logFac

protected LogFaculty logFac

An object to compute and store log(n!)

maxNumNeighbors

protected int maxNumNeighbors

Maximum number of neighboring observations

maxDistNeighbors

protected double maxDistNeighbors

Maximum distance of neighboring observations

Constructor Detail

AssociationDistributionNN

public AssociationDistributionNN(Random rand,
                                 AbstractMultiState<S> Z,
                                 AbstractMultiObservationDistributionIndep<S,T> observationDistrib,
                                 LogProbabilityDensityFunction spatialClutterDistrib,
                                 LogProbabilityDensityFunction spatialNewbornDistrib,
                                 LogProbabilityMassFunction mu,
                                 LogProbabilityMassFunction nu,
                                 double P_D,
                                 int maxNumNeighbors,
                                 double maxDistNeighbors)

Constructor.

Parameters:

rand - random generator for sampling

Z - the current observations

observationDistrib - distribution of the observations model

spatialClutterDistrib - spatial distribution of possible clutter appearance

spatialNewbornDistrib - spatial distribution of possible newborn appearance

mu - distribution of the number of clutter observations

nu - distribution of the number of observations from newborn targets

P_D - probability of target detection

maxNumNeighbors - maximum number of neighboring observations to be considered for each association

maxDistNeighbors - maximum Euklidean distance of neighboring observations to be considered for each association

AssociationDistributionNN

public AssociationDistributionNN(Random rand,
                                 AbstractMultiState<S> Z,
                                 AbstractMultiObservationDistributionIndep<S,T> observationDistrib,
                                 LogProbabilityDensityFunction spatialClutterDistrib,
                                 LogProbabilityDensityFunction spatialNewbornDistrib,
                                 LogProbabilityMassFunction mu,
                                 LogProbabilityMassFunction nu,
                                 double P_D,
                                 int M_max,
                                 int maxNumNeighbors,
                                 double maxDistNeighbors)

Constructor where the maximum number of observations in the time series is specified to avoid some re-computations.

Parameters:

rand - random generator for sampling

Z - the current observations

observationDistrib - distribution of the observations model

spatialClutterDistrib - spatial distribution of possible clutter appearance

spatialNewbornDistrib - spatial distribution of possible newborn appearance

mu - distribution of the number of clutter observations

nu - distribution of the number of observations from newborn targets

P_D - probability of target detection

maxNumNeighbors - maximum number of neighboring observations to be considered for each association

maxDistNeighbors - maximum Euklidean distance of neighboring observations to be considered for each association

M_max - maximum number of observations in the time series

Method Detail

drawSample

public DataAssociation drawSample()

Description copied from interface: SamplingDistribution

Generate a new sample from this density. This method should create a new object.

Specified by:

drawSample in interface SamplingDistribution<DataAssociation>

Overrides:

drawSample in class AssociationDistribution<S extends TargetID,T extends TargetID>

Returns:

new sample object

drawSampleDebug

public DataAssociation drawSampleDebug(DataAssociation groundtruth,
                                       OutputStream ostream)

Overrides:

drawSampleDebug in class AssociationDistribution<S extends TargetID,T extends TargetID>

compute_pczAhead

private double compute_pczAhead(MTBTreeNode pcTree,
                                Vector<Integer> availableTargets,
                                Stack<Integer> observations)

Compute probability of all possible associations of neighboring observations including their likelihood

Parameters:

pcTree -

availableTargets -

observations -

Returns:

compute_qAhead

private MTBTreeNode compute_qAhead(int m,
                                   int mmax,
                                   int kmin,
                                   int bmin,
                                   double logPc_previous,
                                   MTBTreeNode subtree)

Compute tree of data association prior probabilities depending on previous associations.

Parameters:

m -

mmax -

kmin -

bmin -

logPc_previous -

subtree -

Returns:

setNewObservations

public void setNewObservations(AbstractMultiState<S> Z,
                               AbstractMultiObservationDistributionIndep<S,T> observationDistrib)

Overrides:

setNewObservations in class AssociationDistribution<S extends TargetID,T extends TargetID>

kNearestObservations

protected AssociationDistributionNN.ObsDistance[][] kNearestObservations(int maxNumNeighbors,
                                                                         double maxDistNeighbors)

Get the (k) nearest observations for each observation in this.Z, i.e. for each observation z_m all following observations z_{m:M} are sorted by Euklidean distance and stored in an array. If k is larger 0 only the k-nearest observations are returned, if maxDist is larger 0 only observations in that range are returned. Both parameters may be specified.