org.openimaj.image.camera.calibration.CameraCalibrationZhang Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of camera-calibration Show documentation
Show all versions of camera-calibration Show documentation
Camera calibration techniques and associated code
The newest version!
/**
* Copyright (c) 2011, The University of Southampton and the individual contributors.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification,
* are permitted provided that the following conditions are met:
*
* * Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
*
* * Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* * Neither the name of the University of Southampton nor the names of its
* contributors may be used to endorse or promote products derived from this
* software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
* ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
package org.openimaj.image.camera.calibration;
import static java.lang.Math.cos;
import static java.lang.Math.pow;
import static java.lang.Math.sin;
import static java.lang.Math.sqrt;
import java.util.ArrayList;
import java.util.List;
import org.apache.commons.math3.analysis.MultivariateMatrixFunction;
import org.apache.commons.math3.analysis.MultivariateVectorFunction;
import org.apache.commons.math3.fitting.leastsquares.LeastSquaresFactory;
import org.apache.commons.math3.fitting.leastsquares.LeastSquaresOptimizer.Optimum;
import org.apache.commons.math3.fitting.leastsquares.LevenbergMarquardtOptimizer;
import org.apache.commons.math3.fitting.leastsquares.MultivariateJacobianFunction;
import org.apache.commons.math3.linear.ArrayRealVector;
import org.apache.commons.math3.linear.RealVector;
import org.openimaj.citation.annotation.Reference;
import org.openimaj.citation.annotation.ReferenceType;
import org.openimaj.image.camera.Camera;
import org.openimaj.image.camera.CameraIntrinsics;
import org.openimaj.math.geometry.point.Point2d;
import org.openimaj.math.geometry.point.Point3dImpl;
import org.openimaj.math.geometry.transforms.HomographyRefinement;
import org.openimaj.math.geometry.transforms.TransformUtilities;
import org.openimaj.math.matrix.MatrixUtils;
import org.openimaj.util.array.ArrayUtils;
import org.openimaj.util.pair.IndependentPair;
import Jama.Matrix;
/**
* Implementation of Zhengyou Zhang's camera calibration routine using a planar
* calibration pattern. This calibration routine assumes a camera with a 2-term
* radial distortion; the third radial distortion term (k3) and tangential terms
* (p1, p1) of the {@link CameraIntrinsics} will be set to zero.
*
* @author Jonathon Hare ([email protected])
*/
@Reference(
type = ReferenceType.Article,
author = { "Zhengyou Zhang" },
title = "A flexible new technique for camera calibration",
year = "2000",
journal = "Pattern Analysis and Machine Intelligence, IEEE Transactions on",
pages = { "1330", "1334" },
month = "Nov",
number = "11",
volume = "22",
customData = {
"keywords",
"calibration;computer vision;geometry;image sensors;matrix algebra;maximum likelihood estimation;optimisation;3D computer vision;camera calibration;flexible technique;maximum likelihood criterion;planar pattern;radial lens distortion;Calibration;Cameras;Closed-form solution;Computer simulation;Computer vision;Layout;Lenses;Maximum likelihood estimation;Nonlinear distortion;Testing",
"doi", "10.1109/34.888718",
"ISSN", "0162-8828"
})
public class CameraCalibrationZhang {
protected List>> points;
protected List cameras;
/**
* Calibrate a camera using Zhang's method based on the given model-image
* point pairs across a number of images. The model points are in the world
* coordinate system and assumed to be on the Z=0 plane.
*
* @param points
* the pairs of model-image points to calibrate the camera with
* @param width
* the image width of the camera in pixels
* @param height
* the image height of the camera in pixels
*/
public CameraCalibrationZhang(List>> points,
int width, int height)
{
this.points = points;
performCalibration(width, height);
}
protected void performCalibration(int width, int height) {
// compute the homographies
final List homographies = new ArrayList();
for (int i = 0; i < points.size(); i++) {
final List> data = points.get(i);
final Matrix h = HomographyRefinement.SINGLE_IMAGE_TRANSFER.refine(
TransformUtilities.homographyMatrixNorm(data), data);
homographies.add(h);
}
// intial estimate of intrisics and extrinsics
estimateIntrisicAndExtrinsics(homographies, width, height);
// initial estimate of radial distortion
estimateRadialDistortion();
// non-linear optimisation using analytic jacobian
refine();
}
/**
* Get the computed (extrinsic and intrinsic) camera parameters for all
* images used at construction time (in the same order).
*
* @return the camera parameters for each image
*/
public List getCameras() {
return cameras;
}
/**
* Get the computed intrinsic parameters calculated during construction.
*
* @return the intrinsic parameters of the calibrated camera
*/
public CameraIntrinsics getIntrisics() {
return cameras.get(0).intrinsicParameters;
}
private double[] vij(Matrix h, int i, int j) {
h = h.transpose();
final double[] vij = new double[] {
h.get(i, 0) * h.get(j, 0),
h.get(i, 0) * h.get(j, 1) + h.get(i, 1) * h.get(j, 0),
h.get(i, 1) * h.get(j, 1),
h.get(i, 2) * h.get(j, 0) + h.get(i, 0) * h.get(j, 2),
h.get(i, 2) * h.get(j, 1) + h.get(i, 1) * h.get(j, 2),
h.get(i, 2) * h.get(j, 2)
};
return vij;
}
/**
* Compute the initial estimate of the intrinsics
*
* @param homographies
* the homographies
* @param height
* @param width
* @return the intrinsics
*/
private CameraIntrinsics estimateIntrinsics(List homographies, int width, int height) {
final double[][] V = new double[homographies.size() == 2 ? 5 : 2 * homographies.size()][];
for (int i = 0, j = 0; i < homographies.size(); i++, j += 2) {
final Matrix h = homographies.get(i);
V[j] = vij(h, 0, 1); // v12
V[j + 1] = ArrayUtils.subtract(vij(h, 0, 0), vij(h, 1, 1)); // v11-v22
}
if (homographies.size() == 2) {
V[V.length - 1] = new double[] { 0, 1, 0, 0, 0, 0 };
}
final double[] b = MatrixUtils.solveHomogeneousSystem(V);
final double v0 = (b[1] * b[3] - b[0] * b[4]) / (b[0] * b[2] - b[1] * b[1]);
final double lamda = b[5] - (b[3] * b[3] + v0 * (b[1] * b[3] - b[0] * b[4])) / b[0];
final double alpha = Math.sqrt(lamda / b[0]);
final double beta = Math.sqrt(lamda * b[0] / (b[0] * b[2] - b[1] * b[1]));
final double gamma = -b[1] * alpha * alpha * beta / lamda;
final double u0 = gamma * v0 / beta - b[3] * alpha * alpha / lamda;
final Matrix A = new Matrix(new double[][] {
{ alpha, gamma, u0 },
{ 0, beta, v0 },
{ 0, 0, 1 }
});
return new CameraIntrinsics(A, width, height);
}
/**
* Produce an initial estimate of the radial distortion parameters, and
* update the intrinsics
*/
protected void estimateRadialDistortion()
{
final CameraIntrinsics ci = cameras.get(0).intrinsicParameters;
int totalPoints = 0;
for (int i = 0; i < points.size(); i++)
totalPoints += points.get(i).size();
final Matrix D = new Matrix(2 * totalPoints, 2);
final Matrix d = new Matrix(2 * totalPoints, 1);
for (int i = 0, k = 0; i < points.size(); i++) {
final List> pointPairs = points.get(i);
final Matrix idealH = cameras.get(i).computeHomography();
for (int j = 0; j < pointPairs.size(); j++, k++) {
// model point
final Point2d XY = pointPairs.get(j).firstObject();
// transformed ideal point
final Point2d uv = XY.transform(idealH);
// observed point
final Point2d ipt = pointPairs.get(j).secondObject();
d.set(k * 2 + 0, 0, ipt.getX() - uv.getX());
d.set(k * 2 + 1, 0, ipt.getY() - uv.getY());
final double tmp1 = uv.getX() - ci.getPrincipalPointX(); // u-u0
final double tmp2 = uv.getY() - ci.getPrincipalPointY(); // v-v0
final double x = tmp1 / ci.getFocalLengthX(); // (u-u0)/fx
final double y = tmp2 / ci.getFocalLengthY(); // (v-v0)/fy
final double r2 = x * x + y * y;
final double r4 = r2 * r2;
D.set(k * 2 + 0, 0, tmp1 * r2);
D.set(k * 2 + 0, 1, tmp1 * r4);
D.set(k * 2 + 1, 0, tmp2 * r2);
D.set(k * 2 + 1, 1, tmp2 * r4);
}
}
final Matrix result = D.solve(d);
ci.k1 = result.get(0, 0);
ci.k2 = result.get(1, 0);
ci.k3 = 0;
}
/**
* Compute the initial estimate of the intrinsic parameters and then the
* extrinsic parameters assuming zero distortion.
*
* @param homographies
* the homographies
* @param height
* @param width
*/
protected void estimateIntrisicAndExtrinsics(List homographies, int width, int height) {
cameras = new ArrayList(homographies.size());
final CameraIntrinsics intrinsic = estimateIntrinsics(homographies, width, height);
for (int i = 0; i < homographies.size(); i++) {
cameras.add(estimateExtrinsics(homographies.get(i), intrinsic));
}
}
/**
* Estimate the extrinsic parameters for a single camera given its
* homography and intrinsic parameters.
*
* @param h
* the homography
* @param intrinsic
* the intrinsic parameters
* @return the extrinsic parameters
*/
private Camera estimateExtrinsics(Matrix h, CameraIntrinsics intrinsic) {
final Matrix Ainv = intrinsic.calibrationMatrix.inverse();
final Matrix h1 = h.getMatrix(0, 2, 0, 0);
final Matrix h2 = h.getMatrix(0, 2, 1, 1);
final Matrix h3 = h.getMatrix(0, 2, 2, 2);
final Matrix r1 = Ainv.times(h1);
final double lamda = 1 / r1.norm2();
MatrixUtils.times(r1, lamda);
final Matrix r2 = Ainv.times(h2);
MatrixUtils.times(r2, lamda);
final Matrix r3 = new Matrix(new double[][] {
{ r1.get(1, 0) * r2.get(2, 0) - r1.get(2, 0) * r2.get(1, 0) },
{ r1.get(2, 0) * r2.get(0, 0) - r1.get(0, 0) * r2.get(2, 0) },
{ r1.get(0, 0) * r2.get(1, 0) - r1.get(1, 0) * r2.get(0, 0) }
});
final Matrix R = TransformUtilities.approximateRotationMatrix(MatrixUtils.hstack(r1, r2, r3));
final Matrix t = Ainv.times(h3);
MatrixUtils.times(t, lamda);
final Camera ce = new Camera();
ce.intrinsicParameters = intrinsic;
ce.rotation = R;
ce.translation = new Point3dImpl(t.getColumnPackedCopy());
return ce;
}
/**
* This is the implementation of the value function for the optimiser. It
* computes the predicted location of an image point by projecting a model
* point through the camera homography and then applying the distortion. The
* implementation is converted from the C code produced by the following
* matlab symbolic code:
*
*
*
* syms u0 v0 fx fy sk real
* syms tx ty tz wx wy wz real
* syms k1 k2 real
* syms X Y real
*
* % the intrinsic parameter matrix
* K=[fx sk u0; 0 fy v0; 0 0 1];
*
* % Expression for the rotation matrix based on the Rodrigues formula
* theta=sqrt(wx^2+wy^2+wz^2);
* omega=[0 -wz wy; wz 0 -wx; -wy wx 0];
* R = eye(3) + (sin(theta)/theta)*omega + ((1-cos(theta))/theta^2)*(omega*omega);
*
* % Expression for the translation vector
* t=[tx;ty;tz];
*
* % perspective projection of the model point (X,Y)
* uvs=K*[R(:,1) R(:,2) t]*[X; Y; 1];
* u=uvs(1)/uvs(3);
* v=uvs(2)/uvs(3);
*
* % application of 2-term radial distortion
* uu0 = u - u0;
* vv0 = v - v0;
* x = uu0/fx;
* y = vv0/fy;
* r2 = x*x + y*y;
* r4 = r2*r2;
* uv = [u + uu0*(k1*r2 + k2*r4); v + vv0*(k1*r2 + k2*r4)];
* ccode(uv, 'file', 'zhang-value.c')
*
*
*
* @author Jonathon Hare ([email protected])
*
*/
private class Value implements MultivariateVectorFunction {
@Override
public double[] value(double[] params) throws IllegalArgumentException {
int totalPoints = 0;
for (int i = 0; i < points.size(); i++)
totalPoints += points.get(i).size();
final double[] result = new double[2 * totalPoints];
for (int i = 0, k = 0; i < points.size(); i++) {
for (int j = 0; j < points.get(i).size(); j++, k++) {
final double[] tmp = computeValue(i, j, params);
result[k * 2 + 0] = tmp[0];
result[k * 2 + 1] = tmp[1];
}
}
return result;
}
private double[] computeValue(int img, int point, double[] params) {
final double[][] A0 = new double[2][1];
final double X = points.get(img).get(point).firstObject().getX();
final double Y = points.get(img).get(point).firstObject().getY();
final double fx = params[0];
final double fy = params[1];
final double u0 = params[2];
final double v0 = params[3];
final double sk = params[4];
final double k1 = params[5];
final double k2 = params[6];
final double wx = params[img * 6 + 7];
final double wy = params[img * 6 + 8];
final double wz = params[img * 6 + 9];
final double tx = params[img * 6 + 10];
final double ty = params[img * 6 + 11];
final double tz = params[img * 6 + 12];
// begin matlab code
final double t2 = wx * wx;
final double t3 = wy * wy;
final double t4 = wz * wz;
final double t5 = t2 + t3 + t4;
final double t6 = sqrt(t5);
final double t7 = sin(t6);
final double t8 = 1.0 / sqrt(t5);
final double t9 = cos(t6);
final double t10 = t9 - 1.0;
final double t11 = 1.0 / t5;
final double t12 = t7 * t8 * wy;
final double t13 = t10 * t11 * wx * wz;
final double t14 = t12 + t13;
final double t15 = t7 * t8 * wz;
final double t16 = t7 * t8 * wx;
final double t18 = t10 * t11 * wy * wz;
final double t17 = t16 - t18;
final double t19 = Y * t17;
final double t39 = X * t14;
final double t20 = t19 - t39 + tz;
final double t21 = 1.0 / t20;
final double t22 = t10 * t11 * wx * wy;
final double t23 = t3 + t4;
final double t24 = t10 * t11 * t23;
final double t25 = t24 + 1.0;
final double t26 = fx * t25;
final double t27 = t15 + t22;
final double t28 = t17 * u0;
final double t29 = t2 + t4;
final double t30 = t10 * t11 * t29;
final double t31 = t30 + 1.0;
final double t32 = sk * t31;
final double t47 = fx * t27;
final double t33 = t28 + t32 - t47;
final double t34 = Y * t33;
final double t35 = fx * tx;
final double t36 = sk * ty;
final double t37 = tz * u0;
final double t40 = t15 - t22;
final double t43 = sk * t40;
final double t44 = t14 * u0;
final double t45 = t26 + t43 - t44;
final double t46 = X * t45;
final double t48 = t34 + t35 + t36 + t37 + t46;
final double t49 = t21 * t48;
final double t38 = -t49 + u0;
final double t53 = fy * ty;
final double t54 = fy * t40;
final double t55 = t14 * v0;
final double t56 = t54 - t55;
final double t57 = X * t56;
final double t58 = tz * v0;
final double t59 = t17 * v0;
final double t60 = fy * t31;
final double t61 = t59 + t60;
final double t62 = Y * t61;
final double t63 = t53 + t57 + t58 + t62;
final double t64 = t21 * t63;
final double t41 = -t64 + v0;
final double t42 = 1.0 / (fx * fx);
final double t50 = t38 * t38;
final double t51 = t42 * t50;
final double t52 = 1.0 / (fy * fy);
final double t65 = t41 * t41;
final double t66 = t52 * t65;
final double t67 = t51 + t66;
final double t68 = k1 * t67;
final double t69 = t67 * t67;
final double t70 = k2 * t69;
final double t71 = t68 + t70;
A0[0][0] = -t38 * t71 + t21
* (t34 + t35 + t36 + t37 + X * (t26 - t14 * u0 + sk * (t15 - t10 * t11 * wx * wy)));
A0[1][0] = t64 - t41 * t71;
// end matlab code
return new double[] { A0[0][0], A0[1][0] };
}
}
/**
* This is the implementation of the Jacobian function for the optimiser; it
* is the partial derivative of the value function with respect to the
* parameters. The implementation is based on the matlab symbolic code:
*
*
*
* syms u0 v0 fx fy sk real
* syms tx ty tz wx wy wz real
* syms k1 k2 real
* syms X Y real
*
* % the intrinsic parameter matrix
* K=[fx sk u0; 0 fy v0; 0 0 1];
*
* % Expression for the rotation matrix based on the Rodrigues formula
* theta=sqrt(wx^2+wy^2+wz^2);
* omega=[0 -wz wy; wz 0 -wx; -wy wx 0];
* R = eye(3) + (sin(theta)/theta)*omega + ((1-cos(theta))/theta^2)*(omega*omega);
*
* % Expression for the translation vector
* t=[tx;ty;tz];
*
* % perspective projection of the model point (X,Y)
* uvs=K*[R(:,1) R(:,2) t]*[X; Y; 1];
* u=uvs(1)/uvs(3);
* v=uvs(2)/uvs(3);
*
* % application of 2-term radial distortion
* uu0 = u - u0;
* vv0 = v - v0;
* x = uu0/fx;
* y = vv0/fy;
* r2 = x*x + y*y;
* r4 = r2*r2;
* uv = [u + uu0*(k1*r2 + k2*r4); v + vv0*(k1*r2 + k2*r4)];
* J=jacobian(uv,[fx,fy,u0,v0,sk,k1,k2, wx wy wz tx ty tz]);
* ccode(J, 'file', 'zhang-jacobian.c')
*
*
*
* @author Jonathon Hare ([email protected])
*
*/
private class Jacobian implements MultivariateMatrixFunction {
@Override
public double[][] value(double[] params) {
// Note that we're building the jacobian for all cameras/images and
// points. The params vector is 7 + 6*numCameras elements long (7
// intrinsic params and 6 extrinsic per camera)
int totalPoints = 0;
for (int i = 0; i < points.size(); i++)
totalPoints += points.get(i).size();
final double[][] result = new double[2 * totalPoints][];
for (int i = 0, k = 0; i < points.size(); i++) {
for (int j = 0; j < points.get(i).size(); j++, k++) {
final double[][] tmp = computeJacobian(i, j, params);
result[k * 2 + 0] = tmp[0];
result[k * 2 + 1] = tmp[1];
}
}
return result;
}
private double[][] computeJacobian(int img, int point, double[] params) {
final double[][] A0 = new double[2][13];
final double X = points.get(img).get(point).firstObject().getX();
final double Y = points.get(img).get(point).firstObject().getY();
final double fx = params[0];
final double fy = params[1];
final double u0 = params[2];
final double v0 = params[3];
final double sk = params[4];
final double k1 = params[5];
final double k2 = params[6];
final double wx = params[img * 6 + 7];
final double wy = params[img * 6 + 8];
final double wz = params[img * 6 + 9];
final double tx = params[img * 6 + 10];
final double ty = params[img * 6 + 11];
final double tz = params[img * 6 + 12];
// begin matlab code
final double t2 = wx * wx;
final double t3 = wy * wy;
final double t4 = wz * wz;
final double t5 = t2 + t3 + t4;
final double t6 = sqrt(t5);
final double t7 = sin(t6);
final double t8 = 1.0 / sqrt(t5);
final double t9 = cos(t6);
final double t10 = t9 - 1.0;
final double t11 = 1.0 / t5;
final double t12 = t7 * t8 * wy;
final double t13 = t10 * t11 * wx * wz;
final double t14 = t12 + t13;
final double t15 = t7 * t8 * wz;
final double t16 = t7 * t8 * wx;
final double t18 = t10 * t11 * wy * wz;
final double t17 = t16 - t18;
final double t19 = Y * t17;
final double t39 = X * t14;
final double t20 = t19 - t39 + tz;
final double t21 = 1.0 / t20;
final double t22 = t10 * t11 * wx * wy;
final double t23 = t3 + t4;
final double t24 = t10 * t11 * t23;
final double t25 = t24 + 1.0;
final double t26 = fx * t25;
final double t27 = t15 + t22;
final double t28 = t17 * u0;
final double t29 = t2 + t4;
final double t30 = t10 * t11 * t29;
final double t31 = t30 + 1.0;
final double t32 = sk * t31;
final double t45 = fx * t27;
final double t33 = t28 + t32 - t45;
final double t34 = Y * t33;
final double t35 = fx * tx;
final double t36 = sk * ty;
final double t37 = tz * u0;
final double t40 = t15 - t22;
final double t41 = sk * t40;
final double t42 = t14 * u0;
final double t43 = t26 + t41 - t42;
final double t44 = X * t43;
final double t46 = t34 + t35 + t36 + t37 + t44;
final double t47 = t21 * t46;
final double t38 = -t47 + u0;
final double t48 = 1.0 / (fx * fx * fx);
final double t49 = t38 * t38;
final double t50 = t48 * t49 * 2.0;
final double t51 = 1.0 / (fx * fx);
final double t52 = X * t25;
final double t57 = Y * t27;
final double t53 = t52 - t57 + tx;
final double t54 = t21 * t38 * t51 * t53 * 2.0;
final double t55 = t50 + t54;
final double t60 = fy * ty;
final double t61 = fy * t40;
final double t62 = t14 * v0;
final double t63 = t61 - t62;
final double t64 = X * t63;
final double t65 = tz * v0;
final double t66 = t17 * v0;
final double t67 = fy * t31;
final double t68 = t66 + t67;
final double t69 = Y * t68;
final double t70 = t60 + t64 + t65 + t69;
final double t71 = t21 * t70;
final double t56 = -t71 + v0;
final double t58 = t49 * t51;
final double t59 = 1.0 / (fy * fy);
final double t72 = t56 * t56;
final double t73 = t59 * t72;
final double t74 = t58 + t73;
final double t75 = 1.0 / (fy * fy * fy);
final double t76 = t72 * t75 * 2.0;
final double t77 = X * t40;
final double t78 = Y * t31;
final double t79 = t77 + t78 + ty;
final double t80 = t21 * t56 * t59 * t79 * 2.0;
final double t81 = t76 + t80;
final double t82 = k1 * t74;
final double t83 = t74 * t74;
final double t84 = k2 * t83;
final double t85 = t82 + t84;
final double t86 = 1.0 / pow(t5, 3.0 / 2.0);
final double t87 = 1.0 / (t5 * t5);
final double t88 = t9 * t11 * wx * wz;
final double t89 = t2 * t7 * t86 * wy;
final double t90 = t2 * t10 * t87 * wy * 2.0;
final double t91 = t7 * t86 * wx * wy;
final double t92 = t2 * t7 * t86 * wz;
final double t93 = t2 * t10 * t87 * wz * 2.0;
final double t105 = t10 * t11 * wz;
final double t106 = t9 * t11 * wx * wy;
final double t94 = t91 + t92 + t93 - t105 - t106;
final double t95 = t7 * t8;
final double t96 = t2 * t9 * t11;
final double t97 = t10 * t87 * wx * wy * wz * 2.0;
final double t98 = t7 * t86 * wx * wy * wz;
final double t103 = t2 * t7 * t86;
final double t99 = t95 + t96 + t97 + t98 - t103;
final double t100 = t10 * t29 * t87 * wx * 2.0;
final double t101 = t7 * t29 * t86 * wx;
final double t116 = t10 * t11 * wx * 2.0;
final double t102 = t100 + t101 - t116;
final double t104 = t7 * t86 * wx * wz;
final double t107 = X * t94;
final double t108 = Y * t99;
final double t109 = t107 + t108;
final double t110 = 1.0 / (t20 * t20);
final double t111 = t10 * t23 * t87 * wx * 2.0;
final double t112 = t7 * t23 * t86 * wx;
final double t113 = t111 + t112;
final double t117 = t10 * t11 * wy;
final double t114 = t88 + t89 + t90 - t104 - t117;
final double t115 = t94 * u0;
final double t118 = t99 * u0;
final double t119 = fy * t102;
final double t262 = t99 * v0;
final double t120 = t119 - t262;
final double t121 = Y * t120;
final double t122 = fy * t114;
final double t123 = t94 * v0;
final double t124 = t122 + t123;
final double t263 = X * t124;
final double t125 = t121 - t263;
final double t126 = t21 * t125;
final double t127 = t70 * t109 * t110;
final double t128 = t126 + t127;
final double t129 = sk * t114;
final double t141 = fx * t113;
final double t130 = t115 + t129 - t141;
final double t131 = X * t130;
final double t132 = -t88 + t89 + t90 + t104 - t117;
final double t133 = fx * t132;
final double t142 = sk * t102;
final double t134 = t118 + t133 - t142;
final double t135 = Y * t134;
final double t136 = t131 + t135;
final double t137 = t21 * t136;
final double t143 = t46 * t109 * t110;
final double t138 = t137 - t143;
final double t139 = t38 * t51 * t138 * 2.0;
final double t264 = t56 * t59 * t128 * 2.0;
final double t140 = t139 - t264;
final double t144 = t3 * t7 * t86 * wz;
final double t145 = t3 * t10 * t87 * wz * 2.0;
final double t146 = -t91 - t105 + t106 + t144 + t145;
final double t147 = t3 * t7 * t86;
final double t156 = t3 * t9 * t11;
final double t148 = -t95 + t97 + t98 + t147 - t156;
final double t149 = t10 * t29 * t87 * wy * 2.0;
final double t150 = t7 * t29 * t86 * wy;
final double t151 = t149 + t150;
final double t152 = t9 * t11 * wy * wz;
final double t153 = t3 * t7 * t86 * wx;
final double t154 = t3 * t10 * t87 * wx * 2.0;
final double t155 = t7 * t86 * wy * wz;
final double t157 = Y * t146;
final double t158 = X * t148;
final double t159 = t157 + t158;
final double t161 = t10 * t11 * wx;
final double t160 = t152 + t153 + t154 - t155 - t161;
final double t162 = fy * t160;
final double t163 = t148 * v0;
final double t164 = t162 + t163;
final double t165 = X * t164;
final double t166 = fy * t151;
final double t267 = t146 * v0;
final double t167 = t166 - t267;
final double t268 = Y * t167;
final double t168 = t165 - t268;
final double t169 = t21 * t168;
final double t269 = t70 * t110 * t159;
final double t170 = t169 - t269;
final double t171 = t56 * t59 * t170 * 2.0;
final double t172 = -t152 + t153 + t154 + t155 - t161;
final double t173 = fx * t172;
final double t174 = t146 * u0;
final double t189 = sk * t151;
final double t175 = t173 + t174 - t189;
final double t176 = Y * t175;
final double t177 = t10 * t23 * t87 * wy * 2.0;
final double t178 = t7 * t23 * t86 * wy;
final double t190 = t10 * t11 * wy * 2.0;
final double t179 = t177 + t178 - t190;
final double t180 = sk * t160;
final double t181 = t148 * u0;
final double t191 = fx * t179;
final double t182 = t180 + t181 - t191;
final double t183 = X * t182;
final double t184 = t176 + t183;
final double t185 = t21 * t184;
final double t192 = t46 * t110 * t159;
final double t186 = t185 - t192;
final double t187 = t38 * t51 * t186 * 2.0;
final double t188 = t171 + t187;
final double t193 = t4 * t9 * t11;
final double t194 = t4 * t7 * t86 * wx;
final double t195 = t4 * t10 * t87 * wx * 2.0;
final double t196 = -t152 + t155 - t161 + t194 + t195;
final double t197 = t4 * t7 * t86;
final double t198 = t10 * t29 * t87 * wz * 2.0;
final double t199 = t7 * t29 * t86 * wz;
final double t204 = t10 * t11 * wz * 2.0;
final double t200 = t198 + t199 - t204;
final double t201 = t4 * t7 * t86 * wy;
final double t202 = t4 * t10 * t87 * wy * 2.0;
final double t203 = t88 - t104 - t117 + t201 + t202;
final double t205 = t10 * t23 * t87 * wz * 2.0;
final double t206 = t7 * t23 * t86 * wz;
final double t207 = t196 * u0;
final double t208 = t95 + t97 + t98 + t193 - t197;
final double t209 = t203 * u0;
final double t210 = -t95 + t97 + t98 - t193 + t197;
final double t211 = fx * t210;
final double t231 = sk * t200;
final double t212 = t209 + t211 - t231;
final double t213 = Y * t212;
final double t214 = X * t196;
final double t215 = Y * t203;
final double t216 = t214 + t215;
final double t217 = t196 * v0;
final double t218 = fy * t208;
final double t219 = t217 + t218;
final double t220 = X * t219;
final double t221 = fy * t200;
final double t273 = t203 * v0;
final double t222 = t221 - t273;
final double t274 = Y * t222;
final double t223 = t220 - t274;
final double t224 = t21 * t223;
final double t275 = t70 * t110 * t216;
final double t225 = t224 - t275;
final double t226 = t56 * t59 * t225 * 2.0;
final double t227 = -t204 + t205 + t206;
final double t228 = sk * t208;
final double t237 = fx * t227;
final double t229 = t207 + t228 - t237;
final double t230 = X * t229;
final double t232 = t213 + t230;
final double t233 = t21 * t232;
final double t238 = t46 * t110 * t216;
final double t234 = t233 - t238;
final double t235 = t38 * t51 * t234 * 2.0;
final double t236 = t226 + t235;
final double t239 = 1.0 / fx;
final double t240 = 1.0 / fy;
final double t241 = t21 * t56 * t240 * 2.0;
final double t242 = sk * t21 * t38 * t51 * 2.0;
final double t243 = t241 + t242;
final double t244 = t21 * u0;
final double t248 = t46 * t110;
final double t245 = t244 - t248;
final double t246 = t70 * t110;
final double t285 = t21 * v0;
final double t247 = t246 - t285;
final double t249 = t38 * t51 * t245 * 2.0;
final double t286 = t56 * t59 * t247 * 2.0;
final double t250 = t249 - t286;
final double t251 = k1 * t55;
final double t252 = k2 * t55 * t74 * 2.0;
final double t253 = t251 + t252;
final double t254 = k1 * t81;
final double t255 = k2 * t74 * t81 * 2.0;
final double t256 = t254 + t255;
final double t257 = t21 * t79;
final double t258 = t21 * t79 * t85;
final double t259 = k1 * t21 * t38 * t51 * t79 * 2.0;
final double t260 = k2 * t21 * t38 * t51 * t74 * t79 * 4.0;
final double t261 = t259 + t260;
final double t265 = k1 * t140;
final double t266 = k2 * t74 * t140 * 2.0;
final double t270 = k1 * t188;
final double t271 = k2 * t74 * t188 * 2.0;
final double t272 = t270 + t271;
final double t276 = k1 * t236;
final double t277 = k2 * t74 * t236 * 2.0;
final double t278 = t276 + t277;
final double t279 = k1 * t21 * t38 * t239 * 2.0;
final double t280 = k2 * t21 * t38 * t74 * t239 * 4.0;
final double t281 = t279 + t280;
final double t282 = k1 * t243;
final double t283 = k2 * t74 * t243 * 2.0;
final double t284 = t282 + t283;
final double t287 = k1 * t250;
final double t288 = k2 * t74 * t250 * 2.0;
final double t289 = t287 + t288;
A0[0][0] = t21 * t53 + t253
* (u0 - t21 * (t34 + t35 + t36 + t37 + X * (t26 - t14 * u0 + sk * (t15 - t10 * t11 * wx * wy))))
+ t21 * t53 * t85;
A0[0][1] = t38 * t256;
A0[0][2] = 1.0;
A0[0][4] = t257 + t258 + t38 * t261;
A0[0][5] = -t38 * t74;
A0[0][6] = -t38 * t83;
A0[0][7] = t137
- t143
+ t38
* (t265 + t266)
+ t85
* (t21
* (X * (t115 - fx * t113 + sk * (t88 + t89 + t90 - t10 * t11 * wy - t7 * t86 * wx * wz)) + Y
* (t118 + fx * (-t88 + t89 + t90 + t104 - t10 * t11 * wy) - sk * t102)) - t46 * t109
* t110);
A0[0][8] = t185 - t192 + t85 * t186 + t38 * t272;
A0[0][9] = -t238
+ t38
* t278
+ t85
* t234
+ t21
* (t213 + X
* (t207 + sk * (t95 + t97 + t98 + t193 - t4 * t7 * t86) - fx
* (t205 + t206 - t10 * t11 * wz * 2.0)));
A0[0][10] = fx * t21 + t38 * t281 + fx * t21 * t85;
A0[0][11] = sk * t21 + t38 * t284 + sk * t21 * t85;
A0[0][12] = t244 - t46 * t110 + t38 * t289 + t85 * t245;
A0[1][0] = t56 * t253;
A0[1][1] = t257 + t258 + t56 * t256;
A0[1][3] = 1.0;
A0[1][4] = t56 * t261;
A0[1][5] = -t56 * t74;
A0[1][6] = -t56 * t83;
A0[1][7] = -t126 - t127 + t56 * (t265 + t266) - t85 * t128;
A0[1][8] = t169 - t269 + t85 * t170 + t56 * t272;
A0[1][9] = t224 - t275 + t85 * t225 + t56 * t278;
A0[1][10] = t56 * t281;
A0[1][11] = fy * t21 + t56 * t284 + fy * t21 * t85;
A0[1][12] = -t246 + t285 - t85 * t247 + t56 * t289;
// end matlab code
final double[][] result = new double[2][7 + 6 * points.size()];
System.arraycopy(A0[0], 0, result[0], 0, 7);
System.arraycopy(A0[1], 0, result[1], 0, 7);
System.arraycopy(A0[0], 7, result[0], 7 + img * 6, 6);
System.arraycopy(A0[1], 7, result[1], 7 + img * 6, 6);
return result;
}
}
/**
* Stack the observed image locations of the calibration pattern points into
* a vector
*
* @return the observed vector
*/
protected RealVector buildObservedVector()
{
int totalPoints = 0;
for (int i = 0; i < points.size(); i++)
totalPoints += points.get(i).size();
final double[] vec = new double[totalPoints * 2];
for (int i = 0, k = 0; i < points.size(); i++) {
for (int j = 0; j < points.get(i).size(); j++, k++) {
vec[k * 2 + 0] = points.get(i).get(j).secondObject().getX();
vec[k * 2 + 1] = points.get(i).get(j).secondObject().getY();
}
}
return new ArrayRealVector(vec, false);
}
/**
* Perform Levenburg-Marquardt non-linear optimisation to get better
* estimates of the parameters
*/
private void refine()
{
final LevenbergMarquardtOptimizer lm = new LevenbergMarquardtOptimizer();
final RealVector start = buildInitialVector();
final RealVector observed = buildObservedVector();
final int maxEvaluations = 1000;
final int maxIterations = 1000;
final MultivariateVectorFunction value = new Value();
final MultivariateMatrixFunction jacobian = new Jacobian();
final MultivariateJacobianFunction model = LeastSquaresFactory.model(value, jacobian);
final Optimum result = lm.optimize(LeastSquaresFactory.create(model,
observed, start, null, maxEvaluations, maxIterations));
updateEstimates(result.getPoint());
}
/**
* Extract the data from the optimised parameter vector and put it back into
* our camera model
*
* @param point
* the optimised parameter vector
*/
private void updateEstimates(RealVector point) {
final CameraIntrinsics intrinsic = cameras.get(0).intrinsicParameters;
intrinsic.setFocalLengthX(point.getEntry(0));
intrinsic.setFocalLengthY(point.getEntry(1));
intrinsic.setPrincipalPointX(point.getEntry(2));
intrinsic.setPrincipalPointY(point.getEntry(3));
intrinsic.setSkewFactor(point.getEntry(4));
intrinsic.k1 = point.getEntry(5);
intrinsic.k2 = point.getEntry(6);
for (int i = 0; i < cameras.size(); i++) {
final Camera e = cameras.get(i);
final double[] rv = new double[] { point.getEntry(i * 6 + 7), point.getEntry(i * 6 + 8),
point.getEntry(i * 6 + 9) };
e.rotation = TransformUtilities.rodrigues(rv);
e.translation.setX(point.getEntry(i * 6 + 10));
e.translation.setY(point.getEntry(i * 6 + 11));
e.translation.setZ(point.getEntry(i * 6 + 12));
}
}
private RealVector buildInitialVector() {
final CameraIntrinsics intrinsic = cameras.get(0).intrinsicParameters;
final double[] vector = new double[7 + cameras.size() * 6];
vector[0] = intrinsic.getFocalLengthX();
vector[1] = intrinsic.getFocalLengthY();
vector[2] = intrinsic.getPrincipalPointX();
vector[3] = intrinsic.getPrincipalPointY();
vector[4] = intrinsic.getSkewFactor();
vector[5] = intrinsic.k1;
vector[6] = intrinsic.k2;
for (int i = 0; i < cameras.size(); i++) {
final Camera e = cameras.get(i);
final double[] rv = TransformUtilities.rodrigues(e.rotation);
vector[i * 6 + 7] = rv[0];
vector[i * 6 + 8] = rv[1];
vector[i * 6 + 9] = rv[2];
vector[i * 6 + 10] = e.translation.getX();
vector[i * 6 + 11] = e.translation.getY();
vector[i * 6 + 12] = e.translation.getZ();
}
return new ArrayRealVector(vector, false);
}
/**
* Compute the average per-pixel error (in pixels)
*
* @return the average per-pixel error
*/
public double calculateError() {
double error = 0;
int nPoints = 0;
for (int i = 0; i < points.size(); i++) {
for (int j = 0; j < points.get(i).size(); j++) {
nPoints++;
final Point2d model = points.get(i).get(j).firstObject();
final Point2d observed = points.get(i).get(j).secondObject();
final Point2d predicted = cameras.get(i).project(model);
final float dx = observed.getX() - predicted.getX();
final float dy = observed.getY() - predicted.getY();
error += Math.sqrt(dx * dx + dy * dy);
}
}
return error / nPoints;
}
}