Package boofcv.alg.tracker.hybrid

Class HybridTrackerScalePoint<I extends ImageGray<I>,D extends ImageGray<D>,TD extends TupleDesc<TD>>

java.lang.Object

boofcv.alg.tracker.hybrid.HybridTrackerScalePoint<I,D,TD>

public class HybridTrackerScalePoint<I extends ImageGray<I>,D extends ImageGray<D>,TD extends TupleDesc<TD>>
extends Object

Combines a KLT tracker with Detect-Describe-Associate type trackers. Features are nominally tracked using KLT, but after KLT drops a track it is deactivated and made dormant. Upon request, it will attempt to reactivate a dormant track by associating it with newly detected features. After a track has been reactivated it will be tracked normally.

Field Summary

Fields

Modifier and Type	Field	Description
protected AssociateDescriptionSets2D<TD>	associate	Used to associate features using their DDA description
protected FastArray<TD>	detectedDesc
protected FastArray<Point2D_F64>	detectedPixels
protected DogArray_I32	detectedSet
protected DetectDescribePoint<I,TD>	detector	feature detector and describer
protected int	imageHeight
protected int	imageWidth
protected FastArray<TD>	knownDesc
protected FastArray<Point2D_F64>	knownPixels
protected DogArray_I32	knownSet
int	maxInactiveTracks
Random	rand
protected double	toleranceFB
protected long	totalTracks
protected PyramidKltForHybrid<I,D>	trackerKlt	The KLT tracker used to perform the nominal track update
protected FastArray<HybridTrack<TD>>	tracksActive	all active tracks that have been tracked purely by KLT
protected DogArray<HybridTrack<TD>>	tracksAll	List of all tracks (active + inactive) and stores unused track data for future re-use
protected List<HybridTrack<TD>>	tracksDropped	tracks which were dropped this frame
protected FastArray<HybridTrack<TD>>	tracksInactive	tracks not visible in the current frame and were not updated
protected List<HybridTrack<TD>>	tracksSpawned	recently spawned tracks

Constructor Summary

Constructors

Constructor	Description
HybridTrackerScalePoint(PyramidKltForHybrid<I,D> trackerKlt, DetectDescribePoint<I,TD> detector, AssociateDescription2D<TD> associate, int tooCloseRadius)	Configures tracker

Method Summary

Modifier and Type	Method	Description
void	associateInactiveTracks(I input)	These tracks were dropped by KLT at some point in the past.
protected HybridTrack<TD>	createNewTrack()	Creates a new track and sets the descriptor
void	dropAllTracks()	Drops all tracks and recycles the data
void	dropExcessiveInactiveTracks()	If there are too many inactive tracks drop some of them
boolean	dropTrack(HybridTrack<TD> track)	Stops tracking the specified track and recycles its data.
HybridTrack<TD>	dropTrackByAllIndex(int index)	Drops the track using its index in all.
void	pruneActiveTracksWhichAreTooClose()	KLT tracks can drift towards each other (camera moving away) and this is often a bad situation as little new information is added by each track
void	reset()	Sets the tracker into its initial state.
void	spawnNewTracks()	Spawns new tracks from the list of unassociated detections.
void	updateTracks(PyramidDiscrete<I> pyramid, D[] derivX, D[] derivY)	Updates the location and description of tracks using KLT.

Methods inherited from class java.lang.Object

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

Field Details

maxInactiveTracks

public int maxInactiveTracks

rand

public Random rand

imageWidth

protected int imageWidth

imageHeight

protected int imageHeight

trackerKlt

protected PyramidKltForHybrid<I extends ImageGray<I>,D extends ImageGray<D>> trackerKlt

The KLT tracker used to perform the nominal track update

toleranceFB

protected double toleranceFB

detector

protected DetectDescribePoint<I extends ImageGray<I>,TD extends TupleDesc<TD>> detector

feature detector and describer

associate

protected AssociateDescriptionSets2D<TD extends TupleDesc<TD>> associate

Used to associate features using their DDA description

tracksAll

protected DogArray<HybridTrack<TD extends TupleDesc<TD>>> tracksAll

List of all tracks (active + inactive) and stores unused track data for future re-use

tracksActive

protected FastArray<HybridTrack<TD extends TupleDesc<TD>>> tracksActive

all active tracks that have been tracked purely by KLT

tracksInactive

protected FastArray<HybridTrack<TD extends TupleDesc<TD>>> tracksInactive

tracks not visible in the current frame and were not updated

tracksSpawned

protected List<HybridTrack<TD extends TupleDesc<TD>>> tracksSpawned

recently spawned tracks

tracksDropped

protected List<HybridTrack<TD extends TupleDesc<TD>>> tracksDropped

tracks which were dropped this frame

detectedDesc

protected FastArray<TD extends TupleDesc<TD>> detectedDesc

knownDesc

protected FastArray<TD extends TupleDesc<TD>> knownDesc

detectedSet

protected DogArray_I32 detectedSet

knownSet

protected DogArray_I32 knownSet

detectedPixels

protected FastArray<Point2D_F64> detectedPixels

knownPixels

protected FastArray<Point2D_F64> knownPixels

totalTracks

protected long totalTracks

Constructor Details

HybridTrackerScalePoint

public HybridTrackerScalePoint(PyramidKltForHybrid<I,D> trackerKlt,
 DetectDescribePoint<I,TD> detector,
 AssociateDescription2D<TD> associate,
 int tooCloseRadius)

Configures tracker

Parameters:

trackerKlt - KLT tracker used nominally

detector - Feature detector

associate - Association algorithm

tooCloseRadius - If tracks are less than or equal to this distance to each other some will be pruned

Method Details

createNewTrack

protected HybridTrack<TD> createNewTrack()

Creates a new track and sets the descriptor

reset

public void reset()

Sets the tracker into its initial state. Previously declared track data structures are saved for re-use.

updateTracks

public void updateTracks(PyramidDiscrete<I> pyramid,
 D[] derivX,
 D[] derivY)

Updates the location and description of tracks using KLT. Saves a reference to the input image for future processing. Also updates the `totalPureKlt` count.

Parameters:

pyramid - Image pyramid of input.

derivX - Derivative pyramid of input x-axis

derivY - Derivative pyramid of input y-axis

pruneActiveTracksWhichAreTooClose

public void pruneActiveTracksWhichAreTooClose()

KLT tracks can drift towards each other (camera moving away) and this is often a bad situation as little new information is added by each track

associateInactiveTracks

public void associateInactiveTracks(I input)

These tracks were dropped by KLT at some point in the past. See if any of the recent detection match them

spawnNewTracks

public void spawnNewTracks()

Spawns new tracks from the list of unassociated detections. Must call associateInactiveTracks(I) first.

dropExcessiveInactiveTracks

public void dropExcessiveInactiveTracks()

If there are too many inactive tracks drop some of them

dropTrack

public boolean dropTrack(HybridTrack<TD> track)

Stops tracking the specified track and recycles its data.

Parameters:

track - The track being dropped

Returns:

true if the track was being tracked and data was recycled false if not.

dropTrackByAllIndex

public HybridTrack<TD> dropTrackByAllIndex(int index)

Drops the track using its index in all. Note that removeSwap is used so the order will change

Returns:

The dropped track

dropAllTracks

public void dropAllTracks()

Drops all tracks and recycles the data