Package boofcv.alg.sfm.d3

Class VisOdomStereoQuadPnP<T extends ImageGray<T>,TD extends TupleDesc<TD>>

java.lang.Object

boofcv.alg.sfm.d3.VisOdomStereoQuadPnP<T,TD>

All Implemented Interfaces:

VerbosePrint

public class VisOdomStereoQuadPnP<T extends ImageGray<T>,TD extends TupleDesc<TD>>
extends Object
implements VerbosePrint

Stereo visual odometry algorithm which associates image features across two stereo pairs for a total of four images. Image features are first matched between left and right images while applying epipolar constraints. Then the two more recent sets of stereo images are associated with each other in a left to left and right to right fashion. Features which are consistently matched across all four images are saved in a list. RANSAC is then used to remove false positives and estimate camera motion using a PnP type algorithm. Motion is estimated using PNP with RANSAC. The initial 3D location of a feature is found using the stereo pair in the key frame. After the initial motion is found it can optionally be refined. Now that the location of all four cameras is known points are triangulated again using all four views. Then the optional final step is to run bundle adjustment. Features are uniquely tracked from one image to the next. This allows the refined 3D location of each feature to benefit future frames instead of being lost. However, due to the nature of the DDA image tracker, losing track is quite common. The previous stereo pair is referred to as the key frame because it's the reference point that motion is estimated relative to. Inside the code each camera is some times referred to by number. 0 = left camera key frame. 1 = key camera previous frame. 2 = left camera current frame. 3 = right camera current frame.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
class	VisOdomStereoQuadPnP.ImageInfo	Storage for detected features inside an image
static class	VisOdomStereoQuadPnP.QuadMatches	Correspondences between images
static class	VisOdomStereoQuadPnP.TrackQuad	3D coordinate of the feature and its observed location in each image

Field Summary

Fields

Modifier and Type	Field	Description
protected @Nullable PrintStream	profileOut
protected @Nullable PrintStream	verbose

Constructor Summary

Constructors

Constructor	Description
VisOdomStereoQuadPnP(DetectDescribePoint<T,TD> detector, AssociateDescription2D<TD> assocF2F, AssociateDescription2D<TD> assocL2R, Triangulate2ViewsMetric triangulate, ModelMatcher<Se3_F64,Stereo2D3D> matcher, @Nullable ModelFitter<Se3_F64,Stereo2D3D> modelRefiner)	Specifies internal algorithms

Method Summary

Modifier and Type	Method	Description
Se3_F64	getLeftToWorld()
boolean	process(T left, T right)	Estimates camera egomotion from the stereo pair
void	reset()	Resets the algorithm into its original state
void	setCalibration(StereoParameters param)	Sets and saves the stereo camera's calibration
void	setVerbose(@Nullable PrintStream out, @Nullable Set<String> configuration)

Methods inherited from class java.lang.Object

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

Field Details

profileOut

@Nullable
protected @Nullable PrintStream profileOut

verbose

@Nullable
protected @Nullable PrintStream verbose

Constructor Details

VisOdomStereoQuadPnP

public VisOdomStereoQuadPnP(DetectDescribePoint<T,TD> detector,
 AssociateDescription2D<TD> assocF2F,
 AssociateDescription2D<TD> assocL2R,
 Triangulate2ViewsMetric triangulate,
 ModelMatcher<Se3_F64,Stereo2D3D> matcher,
 @Nullable
 @Nullable ModelFitter<Se3_F64,Stereo2D3D> modelRefiner)

Specifies internal algorithms

Parameters:

detector - Estimates image features

assocF2F - Association algorithm used for left to left and right to right

assocL2R - Assocation algorithm used for left to right

triangulate - Used to estimate 3D location of a feature using stereo correspondence

matcher - Robust model estimation. Often RANSAC

modelRefiner - Non-linear refinement of motion estimation

Method Details

setCalibration

public void setCalibration(StereoParameters param)

Sets and saves the stereo camera's calibration

reset

public void reset()

Resets the algorithm into its original state

process

public boolean process(T left,
 T right)

Estimates camera egomotion from the stereo pair

Parameters:

left - Image from left camera

right - Image from right camera

Returns:

true if motion was estimated and false if not

getLeftToWorld

public Se3_F64 getLeftToWorld()

setVerbose

public void setVerbose(@Nullable
 @Nullable PrintStream out,
 @Nullable
 @Nullable Set<String> configuration)

Specified by:

setVerbose in interface VerbosePrint