boofcv.alg.sfm.d3

Class VisOdomPixelDepthPnP<T extends boofcv.struct.image.ImageBase>

java.lang.Object

boofcv.alg.sfm.d3.VisOdomPixelDepthPnP<T>

public class VisOdomPixelDepthPnP<T extends boofcv.struct.image.ImageBase>
extends java.lang.Object

Full 6-DOF visual odometry where a ranging device is assumed for pixels in the primary view and the motion is estimated using a Estimate1ofPnP. Range is usually estimated using stereo cameras, structured light or time of flight sensors. New features are added and removed as needed. Features are removed if they are not part of the inlier feature set for some number of consecutive frames. New features are detected and added if the inlier set falls below a threshold or every turn. Non-linear refinement is optional and appears to provide a very modest improvement in performance. It is recommended that motion is estimated using a P3P algorithm, which is the minimal case. Adding features every frame can be computationally expensive, but having too few features being tracked will degrade accuracy. The algorithm was designed to minimize magic numbers and to be insensitive to small changes in their values. Due to the level of abstraction, it can't take full advantage of the sensors used to estimate 3D feature locations. For example if a stereo camera is used then 3-view geometry can't be used to improve performance.

Constructor Summary

Constructors

Constructor and Description

VisOdomPixelDepthPnP(int thresholdAdd, int thresholdRetire, boolean doublePass, org.ddogleg.fitting.modelset.ModelMatcher<georegression.struct.se.Se3_F64,boofcv.struct.geo.Point2D3D> motionEstimator, ImagePixelTo3D pixelTo3D, boofcv.abst.geo.RefinePnP refine, boofcv.abst.feature.tracker.PointTrackerTwoPass<T> tracker, boofcv.struct.distort.PointTransform_F64 pixelToNorm, boofcv.struct.distort.PointTransform_F64 normToPixel) Configures magic numbers and estimation algorithms.

Method Summary

Methods

Modifier and Type	Method and Description
georegression.struct.se.Se3_F64	getCurrToWorld()
java.util.List<Point2D3DTrack>	getInlierTracks()
org.ddogleg.fitting.modelset.ModelMatcher<georegression.struct.se.Se3_F64,boofcv.struct.geo.Point2D3D>	getMotionEstimator()
long	getTick()
boofcv.abst.feature.tracker.PointTracker<T>	getTracker()
boolean	process(T image) Estimates the motion given the left camera image.
void	reset() Resets the algorithm into its original state
void	setNormToPixel(boofcv.struct.distort.PointTransform_F64 normToPixel)
void	setPixelToNorm(boofcv.struct.distort.PointTransform_F64 pixelToNorm)

Methods inherited from class java.lang.Object

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

Constructor Detail

VisOdomPixelDepthPnP

public VisOdomPixelDepthPnP(int thresholdAdd,
                    int thresholdRetire,
                    boolean doublePass,
                    org.ddogleg.fitting.modelset.ModelMatcher<georegression.struct.se.Se3_F64,boofcv.struct.geo.Point2D3D> motionEstimator,
                    ImagePixelTo3D pixelTo3D,
                    boofcv.abst.geo.RefinePnP refine,
                    boofcv.abst.feature.tracker.PointTrackerTwoPass<T> tracker,
                    boofcv.struct.distort.PointTransform_F64 pixelToNorm,
                    boofcv.struct.distort.PointTransform_F64 normToPixel)

Configures magic numbers and estimation algorithms.

Parameters:

thresholdAdd - Add new tracks when less than this number are in the inlier set. Tracker dependent. Set to a value <= 0 to add features every frame.

thresholdRetire - Discard a track if it is not in the inlier set after this many updates. Try 2

doublePass - Associate image features a second time using the estimated model from the first try to improve results

motionEstimator - PnP motion estimator. P3P algorithm is recommended/

pixelTo3D - Computes the 3D location of pixels.

refine - Optional algorithm for refining the pose estimate. Can be null.

tracker - Point feature tracker.

pixelToNorm - Converts from raw image pixels into normalized image coordinates.

normToPixel - Converts from normalized image coordinates into raw pixels

Method Detail

reset

public void reset()

Resets the algorithm into its original state

process

public boolean process(T image)

Estimates the motion given the left camera image. The latest information required by ImagePixelTo3D should be passed to the class before invoking this function.

Parameters:

image - Camera image.

Returns:

true if successful or false if it failed

getCurrToWorld

public georegression.struct.se.Se3_F64 getCurrToWorld()

getTracker

public boofcv.abst.feature.tracker.PointTracker<T> getTracker()

getMotionEstimator

public org.ddogleg.fitting.modelset.ModelMatcher<georegression.struct.se.Se3_F64,boofcv.struct.geo.Point2D3D> getMotionEstimator()

getInlierTracks

public java.util.List<Point2D3DTrack> getInlierTracks()

setPixelToNorm

public void setPixelToNorm(boofcv.struct.distort.PointTransform_F64 pixelToNorm)

setNormToPixel

public void setNormToPixel(boofcv.struct.distort.PointTransform_F64 normToPixel)

getTick

public long getTick()