Homography¶

class Homography(*args, **kwargs)¶

Bases: ArrayDouble2D

Implementation of an homography and operations on homographies.

This class aims to compute the homography wrt. two images [31] .

The vpHomography class is derived from vpArray2D<double> .

These two images are both described by a set of points. The 2 sets (one per image) are sets of corresponding points : for a point in a image, there is the corresponding point (image of the same 3D point) in the other image points set. These 2 sets are the only data needed to compute the homography. One method used is the one introduced by Ezio Malis during his PhD [26] . A normalization is carried out on this points in order to improve the conditioning of the problem, what leads to improve the stability of the result.

Store and compute the homography such that

\[^a{\bf p} = ^a{\bf H}_b\; ^b{\bf p} \]

with

\[* ^a{\bf H}_b = ^a{\bf R}_b + \frac{^a{\bf t}_b}{^bd} { ^b{\bf n}^T} \]

The tutorial-homography explains how to use this class.

The example below shows also how to manipulate this class to first compute a ground truth homography from camera poses, project pixel coordinates points using an homography and lastly estimate an homography from a subset of 4 matched points in frame a and frame b respectively.

 #include <visp3/core/vpHomogeneousMatrix.h>
 #include <visp3/core/vpMath.h>
 #include <visp3/core/vpMeterPixelConversion.h>
 #include <visp3/vision/vpHomography.h>

 int main()
 {
   // Initialize in the object frame the coordinates in meters of 4 points that
   // belong to a planar object
   vpPoint Po[4];
   Po[0].setWorldCoordinates(-0.1, -0.1, 0);
   Po[1].setWorldCoordinates( 0.2, -0.1, 0);
   Po[2].setWorldCoordinates( 0.1,  0.1, 0);
   Po[3].setWorldCoordinates(-0.1,  0.3, 0);

   // Initialize the pose between camera frame a and object frame o
   vpHomogeneousMatrix aMo(0, 0, 1, 0, 0, 0); // Camera is 1 meter far

   // Initialize the pose between camera frame a and camera frame
   // b. These two frames correspond for example to two successive
   // camera positions
   vpHomogeneousMatrix aMb(0.2, 0.1, 0, 0, 0, vpMath::rad(2));

   // Compute the pose between camera frame b and object frame
   vpHomogeneousMatrix bMo = aMb.inverse() * aMo;

   // Initialize camera intrinsic parameters
   vpCameraParameters cam;

   // Compute the coordinates in pixels of the 4 object points in the
   // camera frame a
   vpPoint Pa[4];
   std::vector<double> xa(4), ya(4); // Coordinates in pixels of the points in frame a
   for(int i=0 ; i < 4 ; ++i) {
     Pa[i] = Po[i]; Pa[i].project(aMo); // Project the points from object frame to camera frame a
     vpMeterPixelConversion::convertPoint(cam,
                                         Pa[i].get_x(), Pa[i].get_y(),
                                         xa[i], ya[i]);
   }

   // Compute the coordinates in pixels of the 4 object points in the
   // camera frame b
   vpPoint Pb[4];
   std::vector<double> xb(4), yb(4); // Coordinates in pixels of the points in frame b
   for(int i=0 ; i < 4 ; ++i) {
     Pb[i] = Po[i]; Pb[i].project(bMo); // Project the points from object frame to camera frame a
   }

   // Compute equation of the 3D plane containing the points in camera frame b
   vpPlane bP(Pb[0], Pb[1], Pb[2]);

   // Compute the corresponding ground truth homography
   vpHomography aHb(aMb, bP);

   std::cout << "Ground truth homography aHb: \n" << aHb<< std::endl;

   // Compute the coordinates of the points in frame b using the ground
   // truth homography and the coordinates of the points in frame a
   vpHomography bHa = aHb.inverse();
   for(int i = 0; i < 4 ; ++i){
     double inv_z = 1. / (bHa[2][0] * xa[i] + bHa[2][1] * ya[i] + bHa[2][2]);

     xb[i] = (bHa[0][0] * xa[i] + bHa[0][1] * ya[i] + bHa[0][2]) * inv_z;
     yb[i] = (bHa[1][0] * xa[i] + bHa[1][1] * ya[i] + bHa[1][2]) * inv_z;
   }

   // Estimate the homography from 4 points coordinates expressed in pixels
   vpHomography::DLT(xb, yb, xa, ya, aHb, true);
   aHb /= aHb[2][2]; // Apply a scale factor to have aHb[2][2] = 1

   std::cout << "Estimated homography aHb: \n" << aHb<< std::endl;
}

Overloaded function.

__init__(self: visp._visp.vision.Homography) -> None

Initialize an homography as identity.

__init__(self: visp._visp.vision.Homography, H: visp._visp.vision.Homography) -> None

Initialize an homography from another homography.

__init__(self: visp._visp.vision.Homography, aMb: visp._visp.core.HomogeneousMatrix, bP: visp._visp.core.Plane) -> None

Construction from translation and rotation and a plane.

__init__(self: visp._visp.vision.Homography, aRb: visp._visp.core.RotationMatrix, atb: visp._visp.core.TranslationVector, bP: visp._visp.core.Plane) -> None

Construction from translation and rotation and a plane.

__init__(self: visp._visp.vision.Homography, tu: visp._visp.core.ThetaUVector, atb: visp._visp.core.TranslationVector, bP: visp._visp.core.Plane) -> None

Construction from translation and rotation and a plane.

__init__(self: visp._visp.vision.Homography, arb: visp._visp.core.PoseVector, bP: visp._visp.core.Plane) -> None

Construction from translation and rotation and a plane.

Methods

`DLT`	From couples of matched points \(^a{\bf p}=(x_a,y_a,1)\) in image a and \(^b{\bf p}=(x_b,y_b,1)\) in image b with homogeneous coordinates, computes the homography matrix by resolving \(^a{\bf p} = ^a{\bf H}_b\; ^b{\bf p}\) using the DLT (Direct Linear Transform) algorithm.
`HLM`	From couples of matched points \(^a{\bf p}=(x_a,y_a,1)\) in image a and \(^b{\bf p}=(x_b,y_b,1)\) in image b with homogeneous coordinates, computes the homography matrix by resolving \(^a{\bf p} = ^a{\bf H}_b\; ^b{\bf p}\) using Ezio Malis linear method (HLM) [25] .
`__init__`	Overloaded function.
`buildFrom`	Overloaded function.
`collineation2homography`	Transform an homography from pixel space to calibrated domain.
`computeDisplacement`	Overloaded function.
`convert`	return: The 3x3 matrix corresponding to the homography.
`det`	Return homography determinant.
`eye`	Set the homography as identity transformation by setting the diagonal to 1 and all other values to 0.
`homography2collineation`	Transform an homography from calibrated domain to pixel space.
`inverse`	Invert the homography.
`load`	Read an homography in a file, verify if it is really an homogeneous matrix.
`project`	Overloaded function.
`projection`	Project the current image point (in frame b) into the frame a using the homography aHb.
`ransac`	From couples of matched points \(^a{\bf p}=(x_a,y_a,1)\) in image a and \(^b{\bf p}=(x_b,y_b,1)\) in image b with homogeneous coordinates, computes the homography matrix by resolving \(^a{\bf p} = ^a{\bf H}_b\; ^b{\bf p}\) using Ransac algorithm.
`resize`	This function is not applicable to an homography that is always a 3-by-3 matrix.
`robust`	From couples of matched points \(^a{\bf p}=(x_a,y_a,1)\) in image a and \(^b{\bf p}=(x_b,y_b,1)\) in image b with homogeneous coordinates, computes the homography matrix by resolving \(^a{\bf p} = ^a{\bf H}_b\; ^b{\bf p}\) using a robust estimation scheme.
`save`	Save an homography in a file.

Inherited Methods

`insert`	Overloaded function.
`getCols`	Return the number of columns of the 2D array.
`insertStatic`	Insert array B in array A at the given position.
`reshape`
`getRows`	Return the number of rows of the 2D array.
`size`	Return the number of elements of the 2D array.
`getMinValue`	Return the array min value.
`getMaxValue`	Return the array max value.
`hadamard`	param m: Second matrix;
`saveYAML`	Save an array in a YAML-formatted file.
`conv2`	Overloaded function.
`t`	Compute the transpose of the array.
`numpy`	Numpy view of the underlying array data.

Operators

`__doc__`
`__init__`	Overloaded function.
`__itruediv__`	Divide all the element of the homography matrix by v : Hij = Hij / v
`__module__`
`__mul__`	Overloaded function.
`__truediv__`	Divide an homography by a scalar.

Attributes

`__annotations__`
`__hash__`

static DLT(xb: list[float], yb: list[float], xa: list[float], ya: list[float], aHb: visp._visp.vision.Homography, normalization: bool = true) → None¶

From couples of matched points \(^a{\bf p}=(x_a,y_a,1)\) in image a and \(^b{\bf p}=(x_b,y_b,1)\) in image b with homogeneous coordinates, computes the homography matrix by resolving \(^a{\bf p} = ^a{\bf H}_b\; ^b{\bf p}\) using the DLT (Direct Linear Transform) algorithm.

At least 4 couples of points are needed.

To do so, we use the DLT algorithm on the data, ie we resolve the linear system by SDV : \(\bf{Ah} =0\) where \(\bf{h}\) is the vector with the terms of \(^a{\bf H}_b\) and \(\mathbf{A}\) depends on the points coordinates.

For each point, in homogeneous coordinates we have:

\[^a{\bf p} = ^a{\bf H}_b\; ^b{\bf p} \]

which is equivalent to:

\[^a{\bf p} \times {^a{\bf H}_b \; ^b{\bf p}} =0 \]

If we note \(\mathbf{h}_j^T\) the \(j^{\textrm{th}}\) line of \(^a{\bf H}_b\) , we can write:

\[\begin{split}^a{\bf H}_b \; ^b{\bf p} = \left( \begin{array}{c}\mathbf{h}_1^T \;^b{\bf p} \\\mathbf{h}_2^T \; ^b{\bf p} \\\mathbf{h}_3^T \;^b{\bf p} \end{array}\right) \end{split}\]

Setting \(^a{\bf p}=(x_{a},y_{a},w_{a})\) , the cross product can be rewritten by:

\[\begin{split}^a{\bf p} \times ^a{\bf H}_b \; ^b{\bf p} =\left( \begin{array}{c}y_{a}\mathbf{h}_3^T \; ^b{\bf p}-w_{a}\mathbf{h}_2^T \; ^b{\bf p} \\w_{a}\mathbf{h}_1^T \; ^b{\bf p} -x_{a}\mathbf{h}_3^T \; ^b{\bf p} \\x_{a}\mathbf{h}_2^T \; ^b{\bf p}- y_{a}\mathbf{h}_1^T \; ^b{\bf p}\end{array}\right) \end{split}\]

\[\begin{split}\underbrace{\left( \begin{array}{ccc}\mathbf{0}^T & -w_{a} \; ^b{\bf p}^T & y_{a} \; ^b{\bf p}^T \\w_{a} \; ^b{\bf p}^T&\mathbf{0}^T & -x_{a} \; ^b{\bf p}^T \\-y_{a} \; ^b{\bf p}^T & x_{a} \; ^b{\bf p}^T & \mathbf{0}^T\end{array}\right)}_{\mathbf{A}_i (3\times 9)} \underbrace{\left( \begin{array}{c}\mathbf{h}_{1}^{T} \\\mathbf{h}_{2}^{T}\\\mathbf{h}_{3}^{T}\end{array}\right)}_{\mathbf{h} (9\times 1)}=0 \end{split}\]

leading to an homogeneous system to be solved: \(\mathbf{A}\mathbf{h}=0\) with \(\mathbf{A}=\left(\mathbf{A}_1^T, ..., \mathbf{A}_i^T, ..., \mathbf{A}_n^T \right)^T\) .

It can be solved using an SVD decomposition:

\[\bf A = UDV^T \]

h is the column of V associated with the smallest singular value of A

Parameters:

xb: list[float]¶: Coordinates vector of matched points in image b. These coordinates are expressed in meters.
yb: list[float]¶: Coordinates vector of matched points in image b. These coordinates are expressed in meters.
xa: list[float]¶: Coordinates vector of matched points in image a. These coordinates are expressed in meters.
ya: list[float]¶: Coordinates vector of matched points in image a. These coordinates are expressed in meters.
aHb: visp._visp.vision.Homography¶: Estimated homography that relies the transformation from image a to image b.
normalization: bool = true¶: When set to true, the coordinates of the points are normalized. The normalization carried out is the one preconized by Hartley.

static HLM(xb: list[float], yb: list[float], xa: list[float], ya: list[float], isplanar: bool, aHb: visp._visp.vision.Homography) → None¶

This method can consider points that are planar or non planar. The algorithm for planar scene implemented in this file is described in Ezio Malis PhD thesis [26] .

If the boolean isplanar is true the points are assumed to be in a plane otherwise there are assumed to be non planar.

Note

See DLT() when the scene is planar.

Parameters:

xb: list[float]¶: Coordinates vector of matched points in image b. These coordinates are expressed in meters.
yb: list[float]¶: Coordinates vector of matched points in image b. These coordinates are expressed in meters.
xa: list[float]¶: Coordinates vector of matched points in image a. These coordinates are expressed in meters.
ya: list[float]¶: Coordinates vector of matched points in image a. These coordinates are expressed in meters.
isplanar: bool¶: If true the points are assumed to be in a plane, otherwise there are assumed to be non planar.
aHb: visp._visp.vision.Homography¶: Estimated homography that relies the transformation from image a to image b.

__eq__(*args, **kwargs)¶

Overloaded function.

__eq__(self: visp._visp.core.ArrayDouble2D, A: visp._visp.core.ArrayDouble2D) -> bool

Equal to comparison operator of a 2D array.

__eq__(self: visp._visp.core.ArrayDouble2D, A: visp._visp.core.ArrayDouble2D) -> bool

Equal to comparison operator of a 2D array.

__eq__(self: visp._visp.core.ArrayDouble2D, A: visp._visp.core.ArrayDouble2D) -> bool

Equal to comparison operator of a 2D array.

__getitem__(*args, **kwargs)¶

Overloaded function.

__getitem__(self: visp._visp.core.ArrayDouble2D, arg0: tuple[int, int]) -> float
__getitem__(self: visp._visp.core.ArrayDouble2D, arg0: int) -> numpy.ndarray[numpy.float64]
__getitem__(self: visp._visp.core.ArrayDouble2D, arg0: slice) -> numpy.ndarray[numpy.float64]
__getitem__(self: visp._visp.core.ArrayDouble2D, arg0: tuple) -> numpy.ndarray[numpy.float64]

__init__(*args, **kwargs)¶

Overloaded function.

__init__(self: visp._visp.vision.Homography) -> None

Initialize an homography as identity.

__init__(self: visp._visp.vision.Homography, H: visp._visp.vision.Homography) -> None

Initialize an homography from another homography.

__init__(self: visp._visp.vision.Homography, aMb: visp._visp.core.HomogeneousMatrix, bP: visp._visp.core.Plane) -> None

Construction from translation and rotation and a plane.

__init__(self: visp._visp.vision.Homography, aRb: visp._visp.core.RotationMatrix, atb: visp._visp.core.TranslationVector, bP: visp._visp.core.Plane) -> None

Construction from translation and rotation and a plane.

__init__(self: visp._visp.vision.Homography, tu: visp._visp.core.ThetaUVector, atb: visp._visp.core.TranslationVector, bP: visp._visp.core.Plane) -> None

Construction from translation and rotation and a plane.

__init__(self: visp._visp.vision.Homography, arb: visp._visp.core.PoseVector, bP: visp._visp.core.Plane) -> None

Construction from translation and rotation and a plane.

__itruediv__(self, v: float) → visp._visp.vision.Homography¶: Divide all the element of the homography matrix by v : Hij = Hij / v

__mul__(*args, **kwargs)¶

Overloaded function.

__mul__(self: visp._visp.vision.Homography, H: visp._visp.vision.Homography) -> visp._visp.vision.Homography

Multiplication by an homography.

vpHomography aHb, bHc;
// Initialize aHb and bHc homographies
vpHomography aHc = aHb * bHc;

Parameters:

H: Homography to multiply with.

__mul__(self: visp._visp.vision.Homography, v: float) -> visp._visp.vision.Homography

Multiply an homography by a scalar.

double v = 1.1;
vpHomography aHb;
// Initialize aHb
vpHomography H = aHb * v;

Parameters:

v: Value of the scalar.

__mul__(self: visp._visp.vision.Homography, b: visp._visp.core.ColVector) -> visp._visp.core.ColVector

Operation a = aHb * b.

Parameters:

b: 3 dimension vector.

__mul__(self: visp._visp.vision.Homography, b_P: visp._visp.core.Point) -> visp._visp.core.Point

From the coordinates of the point in image plane b and the homography between image a and b computes the coordinates of the point in image plane a.

Parameters:

b_P: 2D coordinates of the point in the image plane b.

Returns:

A point with 2D coordinates in the image plane a.

__ne__(*args, **kwargs)¶

Overloaded function.

__ne__(self: visp._visp.core.ArrayDouble2D, A: visp._visp.core.ArrayDouble2D) -> bool

Not equal to comparison operator of a 2D array.

__ne__(self: visp._visp.core.ArrayDouble2D, A: visp._visp.core.ArrayDouble2D) -> bool

Not equal to comparison operator of a 2D array.

__ne__(self: visp._visp.core.ArrayDouble2D, A: visp._visp.core.ArrayDouble2D) -> bool

Not equal to comparison operator of a 2D array.

__truediv__(self, v: float) → visp._visp.vision.Homography¶

Divide an homography by a scalar.

vpHomography aHb;
// Initialize aHb
vpHomography H = aHb / aHb[2][2];

Parameters:

v: float¶: Value of the scalar.

buildFrom(*args, **kwargs)¶

Overloaded function.

buildFrom(self: visp._visp.vision.Homography, aRb: visp._visp.core.RotationMatrix, atb: visp._visp.core.TranslationVector, bP: visp._visp.core.Plane) -> None

Construction from translation and rotation and a plane.

buildFrom(self: visp._visp.vision.Homography, tu: visp._visp.core.ThetaUVector, atb: visp._visp.core.TranslationVector, bP: visp._visp.core.Plane) -> None

Construction from translation and rotation and a plane.

buildFrom(self: visp._visp.vision.Homography, arb: visp._visp.core.PoseVector, bP: visp._visp.core.Plane) -> None

Construction from translation and rotation and a plane.

buildFrom(self: visp._visp.vision.Homography, aMb: visp._visp.core.HomogeneousMatrix, bP: visp._visp.core.Plane) -> None

Construction from homogeneous matrix and a plane.

collineation2homography(self, cam: visp._visp.core.CameraParameters) → visp._visp.vision.Homography¶

Transform an homography from pixel space to calibrated domain.

Note

See homography2collineation()

Parameters:

cam: visp._visp.core.CameraParameters¶

Camera parameters used to fill \({\bf K}\) matrix such as

\[\begin{split}{\bf K} = \left[ \begin{array}{ccc} p_x & 0 & u_0 \\0 & p_y & v_0 \\0 & 0 & 1 \end{array}\right]\end{split}\]

Returns:

The corresponding homography matrix \(\bf H\) in the Euclidean space or calibrated domain.

computeDisplacement(*args, **kwargs)¶

Overloaded function.

computeDisplacement(self: visp._visp.vision.Homography, aRb: visp._visp.core.RotationMatrix, atb: visp._visp.core.TranslationVector, n: visp._visp.core.ColVector) -> None

Compute the camera displacement between two images from the homography \({^a}{\bf H}_b\) which is here an implicit parameter (*this).

Parameters:

aRb: Rotation matrix as an output \({^a}{\bf R}_b\) .
atb: Translation vector as an output \(^a{\bf t}_b\) .
n: Normal vector to the plane as an output.

computeDisplacement(self: visp._visp.vision.Homography, nd: visp._visp.core.ColVector, aRb: visp._visp.core.RotationMatrix, atb: visp._visp.core.TranslationVector, n: visp._visp.core.ColVector) -> None

Compute the camera displacement between two images from the homography \({^a}{\bf H}_b\) which is here an implicit parameter (*this).

Camera displacement between \({^a}{\bf p}\) and \({^a}{\bf p}\) is represented as a rotation matrix \({^a}{\bf R}_b\) and a translation vector \(^a{\bf t}_b\) from which an homogeneous matrix can be build ( vpHomogeneousMatrix ).

Parameters:

nd: Input normal vector to the plane used to compar with the normal vector n extracted from the homography.
aRb: Rotation matrix as an output \({^a}{\bf R}_b\) .
atb: Translation vector as an output \(^a{\bf t}_b\) .
n: Normal vector to the plane as an output.

static conv2(*args, **kwargs)¶

Overloaded function.

conv2(M: visp._visp.core.ArrayDouble2D, kernel: visp._visp.core.ArrayDouble2D, mode: str) -> visp._visp.core.ArrayDouble2D

Perform a 2D convolution similar to Matlab conv2 function: \(M \star kernel\) .

Note

This is a very basic implementation that does not use FFT.

Parameters:

M: First matrix.
kernel: Second matrix.
mode: Convolution mode: “full” (default), “same”, “valid”.

conv2(M: visp._visp.core.ArrayDouble2D, kernel: visp._visp.core.ArrayDouble2D, res: visp._visp.core.ArrayDouble2D, mode: str) -> None

Perform a 2D convolution similar to Matlab conv2 function: \(M \star kernel\) .

Note

This is a very basic implementation that does not use FFT.

Parameters:

M: First array.
kernel: Second array.
res: Result.
mode: Convolution mode: “full” (default), “same”, “valid”.

convert(self) → visp._visp.core.Matrix¶

Returns:: The 3x3 matrix corresponding to the homography.

det(self) → float¶: Return homography determinant.

eye(self) → None¶: Set the homography as identity transformation by setting the diagonal to 1 and all other values to 0.

getCols(self) → int¶: Return the number of columns of the 2D array.

Note

See getRows() , size()

getMaxValue(self) → float¶: Return the array max value.

getMinValue(self) → float¶: Return the array min value.

getRows(self) → int¶: Return the number of rows of the 2D array.

Note

See getCols() , size()

hadamard(self, m: visp._visp.core.ArrayDouble2D) → visp._visp.core.ArrayDouble2D¶

Parameters:

m: visp._visp.core.ArrayDouble2D¶: Second matrix;

Returns:

m1.hadamard(m2) The Hadamard product : \(m1 \circ m2 = (m1 \circ m2)_{i,j} = (m1)_{i,j} (m2)_{i,j}\)

homography2collineation(self, cam: visp._visp.core.CameraParameters) → visp._visp.vision.Homography¶

Transform an homography from calibrated domain to pixel space.

Note

See collineation2homography()

Parameters:

cam: visp._visp.core.CameraParameters¶

Camera parameters used to fill \({\bf K}\) matrix such as

\[\begin{split}{\bf K} = \left[ \begin{array}{ccc} p_x & 0 & u_0 \\0 & p_y & v_0 \\0 & 0 & 1 \end{array}\right]\end{split}\]

Returns:

The corresponding collineation matrix \(\bf G\) in the pixel space.

insert(*args, **kwargs)¶

Overloaded function.

insert(self: visp._visp.core.ArrayDouble2D, A: visp._visp.core.ArrayDouble2D, r: int, c: int) -> None

Insert array A at the given position in the current array.

Warning

Throw vpException::dimensionError if the dimensions of the matrices do not allow the operation.

Parameters:

A: The array to insert.
r: The index of the row to begin to insert data.
c: The index of the column to begin to insert data.

insert(self: visp._visp.core.ArrayDouble2D, A: visp._visp.core.ArrayDouble2D, B: visp._visp.core.ArrayDouble2D, r: int, c: int) -> visp._visp.core.ArrayDouble2D

Insert array B in array A at the given position.

Warning

Throw exception if the sizes of the arrays do not allow the insertion.

Parameters:

A: Main array.
B: Array to insert.
r: Index of the row where to add the array.
c: Index of the column where to add the array.

Returns:

Array with B insert in A.

static insertStatic(A: visp._visp.core.ArrayDouble2D, B: visp._visp.core.ArrayDouble2D, C: visp._visp.core.ArrayDouble2D, r: int, c: int) → None¶

Insert array B in array A at the given position.

Warning

Throw exception if the sizes of the arrays do not allow the insertion.

Parameters:

A: visp._visp.core.ArrayDouble2D¶: Main array.
B: visp._visp.core.ArrayDouble2D¶: Array to insert.
C: visp._visp.core.ArrayDouble2D¶: Result array.
r: int¶: Index of the row where to insert array B.
c: int¶: Index of the column where to insert array B.

inverse(self, bHa: visp._visp.vision.Homography) → None¶

Invert the homography.

Parameters:

bHa: visp._visp.vision.Homography¶: math:bf H^{-1} with H = *this.

load(self: visp._visp.vision.Homography, f: std::basic_ifstream<char, std::char_traits<char> >) → None¶

Read an homography in a file, verify if it is really an homogeneous matrix.

Note

See save()

Parameters:

f: the file. This file has to be written using save() .

numpy(self) → numpy.ndarray[numpy.float64]¶: Numpy view of the underlying array data. This numpy view can be used to directly modify the array.

static project(*args, **kwargs)¶

Overloaded function.

project(cam: visp._visp.core.CameraParameters, bHa: visp._visp.vision.Homography, iPa: visp._visp.core.ImagePoint) -> visp._visp.core.ImagePoint

Given iPa a pixel with coordinates \((u_a,v_a)\) in image a, and the homography bHa in the Euclidean space or calibrated domain that links image a and b, computes the coordinates of the pixel \((u_b,v_b)\) in the image b using the camera parameters matrix \(\bf K\) .

Compute \(^b{\bf p} = {\bf K} \; {^b}{\bf H}_a \; {\bf K}^{-1} {^a}{\bf p}\) with \(^a{\bf p}=(u_a,v_a,1)\) and \(^b{\bf p}=(u_b,v_b,1)\)

Returns:: The coordinates in pixel of the point with coordinates \((u_b,v_b)\) .

project(bHa: visp._visp.vision.Homography, Pa: visp._visp.core.Point) -> visp._visp.core.Point

Given Pa a point with normalized coordinates \((x_a,y_a,1)\) in the image plane a, and the homography bHa in the Euclidean space that links image a and b, computes the normalized coordinates of the point \((x_b,y_b,1)\) in the image plane b.

Compute \(^b{\bf p} = {^b}{\bf H}_a \; {^a}{\bf p}\) with \(^a{\bf p}=(x_a,y_a,1)\) and \(^b{\bf p}=(x_b,y_b,1)\)

Returns:: The coordinates in meter of the point with coordinates \((x_b,y_b)\) .

projection(self, ipb: visp._visp.core.ImagePoint) → visp._visp.core.ImagePoint¶

Project the current image point (in frame b) into the frame a using the homography aHb.

Parameters:

ipb: visp._visp.core.ImagePoint¶: Homography defining the relation between frame a and frame b.

Returns:

The projected image point in the frame a.

static ransac(xb: list[float], yb: list[float], xa: list[float], ya: list[float], aHb: visp._visp.vision.Homography, inliers: list[bool], residual: float, nbInliersConsensus: int, threshold: float, normalization: bool = true) → tuple[bool, list[bool], float]¶

Parameters:

xb: list[float]¶: Coordinates vector of matched points in image b. These coordinates are expressed in meters.
yb: list[float]¶: Coordinates vector of matched points in image b. These coordinates are expressed in meters.
xa: list[float]¶: Coordinates vector of matched points in image a. These coordinates are expressed in meters.
ya: list[float]¶: Coordinates vector of matched points in image a. These coordinates are expressed in meters.
aHb: visp._visp.vision.Homography¶: Estimated homography that relies the transformation from image a to image b.
inliers: list[bool]¶: Vector that indicates if a matched point is an inlier (true) or an outlier (false).
residual: float¶: Global residual computed as \(r = \sqrt{1/n \sum_{inliers} {\| {^a{\bf p} - {\hat{^a{\bf H}_b}} {^b{\bf p}}} \|}^{2}}\) with \(n\) the number of inliers.
nbInliersConsensus: int¶: Minimal number of points requested to fit the estimated homography.
threshold: float¶: Threshold for outlier removing. A point is considered as an outlier if the reprojection error \(\| {^a{\bf p} - {\hat{^a{\bf H}_b}} {^b{\bf p}}} \|\) is greater than this threshold.
normalization: bool = true¶: When set to true, the coordinates of the points are normalized. The normalization carried out is the one preconized by Hartley.

Returns:

A tuple containing:

true if the homography could be computed, false otherwise.
inliers: Vector that indicates if a matched point is an inlier (true) or an outlier (false).
residual: Global residual computed as \(r = \sqrt{1/n \sum_{inliers} {\| {^a{\bf p} - {\hat{^a{\bf H}_b}} {^b{\bf p}}} \|}^{2}}\) with \(n\) the number of inliers.

reshape(self, nrows: int, ncols: int) → None¶

resize(self, nrows: int, ncols: int, flagNullify: bool = true) → None¶: This function is not applicable to an homography that is always a 3-by-3 matrix.

static robust(xb: list[float], yb: list[float], xa: list[float], ya: list[float], aHb: visp._visp.vision.Homography, inliers: list[bool], residual: float, weights_threshold: float = 0.4, niter: int = 4, normalization: bool = true) → tuple[list[bool], float]¶

This method is to compare to DLT() except that here a robust estimator is used to reject couples of points that are considered as outliers.

At least 4 couples of points are needed.

Note

See DLT() , ransac()

Parameters:

xb: list[float]¶: Coordinates vector of matched points in image b. These coordinates are expressed in meters.
yb: list[float]¶: Coordinates vector of matched points in image b. These coordinates are expressed in meters.
xa: list[float]¶: Coordinates vector of matched points in image a. These coordinates are expressed in meters.
ya: list[float]¶: Coordinates vector of matched points in image a. These coordinates are expressed in meters.
aHb: visp._visp.vision.Homography¶: Estimated homography that relies the transformation from image a to image b.
inliers: list[bool]¶: Vector that indicates if a matched point is an inlier (true) or an outlier (false).
residual: float¶: Global residual computed as \(r = \sqrt{1/n \sum_{inliers} {\| {^a{\bf p} - {\hat{^a{\bf H}_b}} {^b{\bf p}}} \|}^{2}}\) with \(n\) the number of inliers.
weights_threshold: float = 0.4¶: Threshold applied on the weights updated during the robust estimation and used to consider if a point is an outlier or an inlier. Values should be in [0:1]. A couple of matched points that have a weight lower than this threshold is considered as an outlier. A value equal to zero indicates that all the points are inliers.
niter: int = 4¶: Number of iterations of the estimation process.
normalization: bool = true¶: When set to true, the coordinates of the points are normalized. The normalization carried out is the one preconized by Hartley.

Returns:

A tuple containing:

inliers: Vector that indicates if a matched point is an inlier (true) or an outlier (false).
residual: Global residual computed as \(r = \sqrt{1/n \sum_{inliers} {\| {^a{\bf p} - {\hat{^a{\bf H}_b}} {^b{\bf p}}} \|}^{2}}\) with \(n\) the number of inliers.

save(self: visp._visp.vision.Homography, f: std::basic_ofstream<char, std::char_traits<char> >) → None¶: Save an homography in a file. The load() function allows then to read and set the homography from this file.

Note

See load()

static saveYAML(filename: str, A: visp._visp.core.ArrayDouble2D, header: str =) → bool¶

Save an array in a YAML-formatted file.

Here is an example of outputs.

vpArray2D<double> M(3,4);
vpArray2D::saveYAML("matrix.yml", M, "example: a YAML-formatted header");
vpArray2D::saveYAML("matrixIndent.yml", M, "example:\n    - a YAML-formatted \
header\n    - with inner indentation");

Content of matrix.yml:

example: a YAML-formatted header
rows: 3
cols: 4
data:
  - [0, 0, 0, 0]
  - [0, 0, 0, 0]
  - [0, 0, 0, 0]

Content of matrixIndent.yml:

example:
    - a YAML-formatted header
    - with inner indentation
rows: 3
cols: 4
data:
    - [0, 0, 0, 0]
    - [0, 0, 0, 0]
    - [0, 0, 0, 0]

Note

See loadYAML()

Parameters:

filename: absolute file name.
A: array to be saved in the file.
header: optional lines that will be saved at the beginning of the file. Should be YAML-formatted and will adapt to the indentation if any.

Returns:

Returns true if success.

size(self) → int¶: Return the number of elements of the 2D array.

t(self) → visp._visp.core.ArrayDouble2D¶

Compute the transpose of the array.

Returns:: vpArray2D<Type> C = A^T

__hash__ = None¶