codeanticode.progpuklt
Class FeatureTrackerParameters

java.lang.Object
  codeanticode.progpuklt.FeatureTrackerParameters

public class FeatureTrackerParameters
extends java.lang.Object

This class encapsulates the parameters of the GPU-KLT algorithm.

Author:

Andres Colubri

Field Summary

float	convergenceThreshold If the position update after the inner iterations (after nIterations steps) has not converged (i.e.
float	detectBorderMargin Suppress corners very close to the image borders.
boolean	explicitLod Enables explicit LOD in the pyramid generation (w/out derivatives).
int	kernelSize Kernel size used in the pyramid with derivatives generation.
int	levelSkip After getting an initial estimate of the new feature position in the previous pyramid level, the next level used for refinement is level+levelSkip.
float	minCornerness The minimum value of the cornerness required for a pixel to classify as corner.
int	minDistance Suppress corners/features that are two close.
int	nIterations Number of updates on the position to solve the actually non-linear minimization problem (i.e.
int	nLevels Number of pyramid levels generated and used for tracking.
int	nTrackedFrames Interval for feature re-detection.
float	SSDThreshold If the error function we are minimizing (with respect to the position update) is not small enough (i.e.
float	trackBorderMargin Features very close to the image border are discarded.
boolean	trackWithGain Should simultaneous gain estimation be enabled (allows some variation in brightness in the images) or not (faster tracking).
int	windowWidth Size of the window around the feature used for the position update.

Constructor Summary

FeatureTrackerParameters()
Default constructor of the class.

FeatureTrackerParameters(int nLevels, boolean trackWithGain)
This constructor allows to set the number nLevels of levels and disable/enable gain tracking.

Method Summary

Methods inherited from class java.lang.Object

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

Field Detail

nIterations

public int nIterations

Number of updates on the position to solve the actually non-linear minimization problem (i.e. number of inner iterations).

nLevels

public int nLevels

Number of pyramid levels generated and used for tracking.

levelSkip

public int levelSkip

After getting an initial estimate of the new feature position in the previous pyramid level, the next level used for refinement is level+levelSkip. reasonable values are 1 (all levels) and nLevels-1 (2 levels). With levelSkip equal to 1 all pyramid levels are fully evaluated; levelSkip = nLevels-1, only the coarsest and the full resolution level are used.

windowWidth

public int windowWidth

Size of the window around the feature used for the position update. The used window is windowWidth x windowWidth with the feature in the center.

trackBorderMargin

public float trackBorderMargin

Features very close to the image border are discarded. You want to have at least trackBorderMargin>=windowWidth/2 in order to have the feature completely within the image.

convergenceThreshold

public float convergenceThreshold

If the position update after the inner iterations (after nIterations steps) has not converged (i.e. the length of the update > convergenceThreshold in the final iteration), discard the feature. It is measured in pixels.

SSDThreshold

public float SSDThreshold

If the error function we are minimizing (with respect to the position update) is not small enough (i.e. >SSDThreshold), discard the feature (no similarly looking patch found).

trackWithGain

public boolean trackWithGain

Should simultaneous gain estimation be enabled (allows some variation in brightness in the images) or not (faster tracking).

minDistance

public int minDistance

Suppress corners/features that are two close. If two corners have distance < minDistance, then the weaker one is suppressed. This enables a reasonable uniform distribution of features in the image.

minCornerness

public float minCornerness

The minimum value of the cornerness required for a pixel to classify as corner. Highly textured areas have larger cornerness.

detectBorderMargin

public float detectBorderMargin

Suppress corners very close to the image borders. Setting detectBorderMargin = trackBorderMargin is fine.

kernelSize

public int kernelSize

Kernel size used in the pyramid with derivatives generation.

explicitLod

public boolean explicitLod

Enables explicit LOD in the pyramid generation (w/out derivatives).

nTrackedFrames

public int nTrackedFrames

Interval for feature re-detection.

Constructor Detail

FeatureTrackerParameters

public FeatureTrackerParameters()

Default constructor of the class. Sets the parameters to their default values.

FeatureTrackerParameters

public FeatureTrackerParameters(int nLevels,
                                boolean trackWithGain)

This constructor allows to set the number nLevels of levels and disable/enable gain tracking.