net.sf.jaer.stereopsis
Class DisparityFilter

java.lang.Object
  net.sf.jaer.eventprocessing.EventFilter
      net.sf.jaer.eventprocessing.EventFilter2D
          net.sf.jaer.stereopsis.DisparityFilter

All Implemented Interfaces:

java.util.Observer, FrameAnnotater

public class DisparityFilter
extends EventFilter2D
implements FrameAnnotater, java.util.Observer

The filter calculates x-coordinate disparities for every single event, by averaging over disparities of previous events in the local neighborhood. Therefore the disparites of single events should be spatially and temporally smooth. First, the algorithm calculates a weighted average of previous disparity values in the neighborhood (= oldMeanDisp). The weight depends on the time difference between the actual and the previous events, s.t. recent events will have more weight. Depending on the mean time difference from the previous events, the search range for the actual stereomatching will be restricted around oldMeanDisp. The disparity of the actual event is then calculated by a linear interpolation between the best match in the restricted search range and oldMeanDisp. Because looking at all points in a certain radius is too expansive for neighborhood computation, only a sparse set of surrounding points will be evaluated. These points are stored in the file 'neighbors.dat'. New sets of points can be generated easyly by using the matlab 'function neighborhoodCreator.m'.

Author:

Peter Hess

Field Summary

Fields inherited from class net.sf.jaer.eventprocessing.EventFilter2D

enclosedFilter, out

Fields inherited from class net.sf.jaer.eventprocessing.EventFilter

annotationEnabled, chip, enclosedFilterChain, filterEnabled, log, perf, propertyTooltipMap, support

Constructor Summary

DisparityFilter(AEChip chip)
Creates a new instance of GlobalXDisparityFilter3

Method Summary

void	annotate(float[][][] frame) annotate the RGB frame somehow by color pixels in the rendered pixel frame data.
void	annotate(javax.media.opengl.GLAutoDrawable drawable) Each annotator enters annotate with graphics context current, in coordinates with pixel 0,0 in LL corner (note opposite from Java2D) and pixel spacing 1 unit after the scaling transform (which is already active).
void	annotate(java.awt.Graphics2D g) each annotator is called by the relevant class (e.g.
EventPacket	filterPacket(EventPacket in) Subclasses implement this method to define custom processing.
int	getDisparity()
float	getDistFactor()
java.lang.Object	getFilterState() should return the filter state in some useful form
int	getMaxDisp()
float	getRangeFactor()
float	getSmoothFactor()
void	initFilter() this should allocate and initialize memory: it may be called when the chip e.g.
boolean	isGeneratingFilter()
void	resetFilter() should reset the filter to initial state
void	setDistFactor(float distFactor) Set the scaling factor for the event correlation depending on time difference.
void	setMaxDisp(int maxDisp) Maximal disparity which is considered for event matching.
void	setRangeFactor(float rf)
void	setSmoothFactor(float smoothFactor) This factor weights the influence of disparities from neigbouring events.
void	update(java.util.Observable o, java.lang.Object arg)

Methods inherited from class net.sf.jaer.eventprocessing.EventFilter2D

checkOutputPacketEventType, checkOutputPacketEventType, getEnclosedFilter, resetOut, setEnclosedFilter, setFilterEnabled

Methods inherited from class net.sf.jaer.eventprocessing.EventFilter

getChip, getDescription, getEnclosedFilterChain, getEnclosingFilter, getPrefs, getPropertyChangeSupport, getPropertyTooltip, isAnnotationEnabled, isEnclosed, isFilterEnabled, prefsEnabledKey, setAnnotationEnabled, setChip, setEnclosed, setEnclosedFilter, setEnclosedFilterChain, setEnclosingFilter, setPreferredEnabledState, setPrefs, setPropertyTooltip

Methods inherited from class java.lang.Object

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

Methods inherited from interface net.sf.jaer.graphics.FrameAnnotater

isAnnotationEnabled, setAnnotationEnabled

Constructor Detail

DisparityFilter

public DisparityFilter(AEChip chip)

Creates a new instance of GlobalXDisparityFilter3

Method Detail

initFilter

public void initFilter()

Description copied from class: EventFilter

this should allocate and initialize memory: it may be called when the chip e.g. size parameters are changed after creation of the filter

Specified by:

initFilter in class EventFilter

resetFilter

public void resetFilter()

Description copied from class: EventFilter

should reset the filter to initial state

Specified by:

resetFilter in class EventFilter

getFilterState

public java.lang.Object getFilterState()

Description copied from class: EventFilter

should return the filter state in some useful form

Specified by:

getFilterState in class EventFilter

isGeneratingFilter

public boolean isGeneratingFilter()

update

public void update(java.util.Observable o,
                   java.lang.Object arg)

Specified by:

update in interface java.util.Observer

setDistFactor

public void setDistFactor(float distFactor)

Set the scaling factor for the event correlation depending on time difference.

getDistFactor

public float getDistFactor()

setRangeFactor

public void setRangeFactor(float rf)

getRangeFactor

public float getRangeFactor()

setSmoothFactor

public void setSmoothFactor(float smoothFactor)

This factor weights the influence of disparities from neigbouring events. A higher value will result in more smoothing, but reaction to changing object distance will also be slower.

getSmoothFactor

public float getSmoothFactor()

setMaxDisp

public void setMaxDisp(int maxDisp)

Maximal disparity which is considered for event matching.

getMaxDisp

public int getMaxDisp()

getDisparity

public int getDisparity()

filterPacket

public EventPacket filterPacket(EventPacket in)

Description copied from class: EventFilter2D

Subclasses implement this method to define custom processing.

Specified by:

filterPacket in class EventFilter2D

Parameters:

in - the input packet

Returns:

the output packet

annotate

public void annotate(java.awt.Graphics2D g)

Description copied from interface: FrameAnnotater

each annotator is called by the relevant class (e.g. EyeTracker) and enters annotate with graphics context current, in coordinates with pixel 0,0 in UL corner and pixel spacing 1 unit before scaling transform (which is already active).

Specified by:

annotate in interface FrameAnnotater

Parameters:

g - the Graphics2D context

annotate

public void annotate(float[][][] frame)

Description copied from interface: FrameAnnotater

annotate the RGB frame somehow by color pixels in the rendered pixel frame data.

Specified by:

annotate in interface FrameAnnotater

Parameters:

frame - the RGB pixel information. First dimension is Y, second is X, third is RGB

annotate

public void annotate(javax.media.opengl.GLAutoDrawable drawable)

Description copied from interface: FrameAnnotater

Each annotator enters annotate with graphics context current, in coordinates with pixel 0,0 in LL corner (note opposite from Java2D) and pixel spacing 1 unit after the scaling transform (which is already active). The FrameAnnotater then can use JOGL calls to render to the screen by getting the GL context, e.g. the following code, used in the context of an AEChip object, draws a golden lines from LL to UR of the pixel array.

        GL gl = drawable.getGL();
        gl.glBegin(GL.GL_LINES);
        gl.glColor3f(.5f, .5f, 0);
        gl.glVertex2f(0, 0);
        gl.glVertex2f(getSizeX() - 1, getSizeY() - 1);
        gl.glEnd();
 

Specified by:

annotate in interface FrameAnnotater

Parameters:

drawable - the OpenGL drawable components, e.g., GLCanvas