boofcv.alg.flow

Class HornSchunckPyramid<T extends boofcv.struct.image.ImageSingleBand>

java.lang.Object

boofcv.alg.flow.DenseFlowPyramidBase<T>

boofcv.alg.flow.HornSchunckPyramid<T>

public class HornSchunckPyramid<T extends boofcv.struct.image.ImageSingleBand>
extends DenseFlowPyramidBase<T>

Pyramidal implementation of Horn-Schunck based on the discussion in [1]. The problem formulation has been modified from the original found in [2] to account for larger displacements. The Euler-Lagrange equations are solved using Successive Over-Relaxation (SOR).

1. Meinhardt-Llopis, Enric and Sánchez Pérez, Javier and Kondermann, Daniel, "Horn-Schunck Optical Flow with a Multi-Scale Strategy" vol 3, 2013, Image Processing On Line
2. Horn, Berthold K., and Brian G. Schunck. "Determining optical flow." 1981 Technical Symposium East. International Society for Optics and Photonics, 1981.

Field Summary

Fields

Modifier and Type	Field and Description
protected boofcv.struct.image.ImageFloat32	flowX
protected boofcv.struct.image.ImageFloat32	flowY
protected boofcv.struct.image.ImageFloat32	initFlowX
protected boofcv.struct.image.ImageFloat32	initFlowY
protected boofcv.struct.image.ImageFloat32	warpDeriv2X
protected boofcv.struct.image.ImageFloat32	warpDeriv2Y
protected boofcv.struct.image.ImageFloat32	warpImage2

Fields inherited from class boofcv.alg.flow.DenseFlowPyramidBase

interp, pyr1, pyr2

Constructor Summary

Constructors

Constructor and Description
HornSchunckPyramid(ConfigHornSchunckPyramid config, boofcv.alg.interpolate.InterpolatePixelS<boofcv.struct.image.ImageFloat32> interp) Configures flow estimation

Method Summary

Methods

Modifier and Type	Method and Description
protected static float	A_safe(int x, int y, boofcv.struct.image.ImageFloat32 flow) See equation 25.
protected static float	A(int x, int y, boofcv.struct.image.ImageFloat32 flow) See equation 25.
boofcv.struct.image.ImageFloat32	getFlowX()
boofcv.struct.image.ImageFloat32	getFlowY()
protected void	interpolateFlowScale(boofcv.struct.image.ImageFloat32 prev, boofcv.struct.image.ImageFloat32 curr) Takes the flow from the previous lower resolution layer and uses it to initialize the flow in the current layer.
protected void	interpolateFlowScale(int widthNew, int heightNew) Provides an initial estimate for the flow by interpolating values from the previous layer.
void	process(boofcv.struct.pyramid.ImagePyramid<boofcv.struct.image.ImageFloat32> image1, boofcv.struct.pyramid.ImagePyramid<boofcv.struct.image.ImageFloat32> image2) Computes dense optical flow from the provided image pyramid.
protected void	processLayer(boofcv.struct.image.ImageFloat32 image1, boofcv.struct.image.ImageFloat32 image2, boofcv.struct.image.ImageFloat32 derivX2, boofcv.struct.image.ImageFloat32 derivY2) Computes the flow for a layer using Taylor series expansion and Successive Over-Relaxation linear solver.
protected static float	safe(int x, int y, boofcv.struct.image.ImageFloat32 image) Ensures pixel values are inside the image.
protected void	warpImageTaylor(boofcv.struct.image.ImageFloat32 before, boofcv.struct.image.ImageFloat32 flowX, boofcv.struct.image.ImageFloat32 flowY, boofcv.struct.image.ImageFloat32 after) Takes the flow from the previous lower resolution layer and uses it to initialize the flow in the current layer.

Methods inherited from class boofcv.alg.flow.DenseFlowPyramidBase

imageNormalization, process

Methods inherited from class java.lang.Object

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

Field Detail

flowX

protected boofcv.struct.image.ImageFloat32 flowX

flowY

protected boofcv.struct.image.ImageFloat32 flowY

initFlowX

protected boofcv.struct.image.ImageFloat32 initFlowX

initFlowY

protected boofcv.struct.image.ImageFloat32 initFlowY

warpImage2

protected boofcv.struct.image.ImageFloat32 warpImage2

warpDeriv2X

protected boofcv.struct.image.ImageFloat32 warpDeriv2X

warpDeriv2Y

protected boofcv.struct.image.ImageFloat32 warpDeriv2Y

Constructor Detail

HornSchunckPyramid

public HornSchunckPyramid(ConfigHornSchunckPyramid config,
                  boofcv.alg.interpolate.InterpolatePixelS<boofcv.struct.image.ImageFloat32> interp)

Configures flow estimation

Parameters:

config - Configuration parameters

interp - Interpolation for image flow between image layers and warping. Overrides selection in config.

Method Detail

process

public void process(boofcv.struct.pyramid.ImagePyramid<boofcv.struct.image.ImageFloat32> image1,
           boofcv.struct.pyramid.ImagePyramid<boofcv.struct.image.ImageFloat32> image2)

Computes dense optical flow from the provided image pyramid. Image gradient for each layer should be computed directly from the layer images.

Specified by:

process in class DenseFlowPyramidBase<T extends boofcv.struct.image.ImageSingleBand>

Parameters:

image1 - Pyramid of first image

image2 - Pyramid of second image

interpolateFlowScale

protected void interpolateFlowScale(int widthNew,
                        int heightNew)

Provides an initial estimate for the flow by interpolating values from the previous layer.

interpolateFlowScale

protected void interpolateFlowScale(boofcv.struct.image.ImageFloat32 prev,
                        boofcv.struct.image.ImageFloat32 curr)

Takes the flow from the previous lower resolution layer and uses it to initialize the flow in the current layer. Adjusts for change in image scale.

Overrides:

interpolateFlowScale in class DenseFlowPyramidBase<T extends boofcv.struct.image.ImageSingleBand>

warpImageTaylor

protected void warpImageTaylor(boofcv.struct.image.ImageFloat32 before,
                   boofcv.struct.image.ImageFloat32 flowX,
                   boofcv.struct.image.ImageFloat32 flowY,
                   boofcv.struct.image.ImageFloat32 after)

Takes the flow from the previous lower resolution layer and uses it to initialize the flow in the current layer. Adjusts for change in image scale.

Overrides:

warpImageTaylor in class DenseFlowPyramidBase<T extends boofcv.struct.image.ImageSingleBand>

processLayer

protected void processLayer(boofcv.struct.image.ImageFloat32 image1,
                boofcv.struct.image.ImageFloat32 image2,
                boofcv.struct.image.ImageFloat32 derivX2,
                boofcv.struct.image.ImageFloat32 derivY2)

Computes the flow for a layer using Taylor series expansion and Successive Over-Relaxation linear solver. Flow estimates from previous layers are feed into this by setting initFlow and flow to their values.

A_safe

protected static float A_safe(int x,
           int y,
           boofcv.struct.image.ImageFloat32 flow)

See equation 25. Safe version

A

protected static float A(int x,
      int y,
      boofcv.struct.image.ImageFloat32 flow)

See equation 25. Fast unsafe version

safe

protected static float safe(int x,
         int y,
         boofcv.struct.image.ImageFloat32 image)

Ensures pixel values are inside the image. If output it is assigned to the nearest pixel inside the image

getFlowX

public boofcv.struct.image.ImageFloat32 getFlowX()

getFlowY

public boofcv.struct.image.ImageFloat32 getFlowY()