InhomogeneousRightEpipoleRefiner.java
/*
* Copyright (C) 2017 Alberto Irurueta Carro (alberto@irurueta.com)
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.irurueta.ar.epipolar.refiners;
import com.irurueta.algebra.Matrix;
import com.irurueta.ar.epipolar.FundamentalMatrix;
import com.irurueta.ar.epipolar.InvalidFundamentalMatrixException;
import com.irurueta.geometry.CoordinatesType;
import com.irurueta.geometry.InhomogeneousPoint2D;
import com.irurueta.geometry.Point2D;
import com.irurueta.geometry.Transformation2D;
import com.irurueta.geometry.estimators.LockedException;
import com.irurueta.geometry.estimators.NotReadyException;
import com.irurueta.geometry.refiners.RefinerException;
import com.irurueta.numerical.EvaluationException;
import com.irurueta.numerical.GradientEstimator;
import com.irurueta.numerical.fitting.LevenbergMarquardtMultiDimensionFitter;
import com.irurueta.numerical.fitting.LevenbergMarquardtMultiDimensionFunctionEvaluator;
import com.irurueta.numerical.robust.InliersData;
import java.util.BitSet;
import java.util.List;
/**
* Refines the epipole of a fundamental matrix formed by an initial epipole
* estimation and an estimated homography.
* Any fundamental matrix can be expressed as F = [e']x*H, where
* e' is the epipole on the right view, []x is the skew matrix, and H is a
* non-degenerate homography.
* This class refines an initial epipole so that residuals from provided point
* correspondences generating fundamental matrix F are reduced.
* This class is especially useful in cases where geometry of the scene is
* degenerate (e.g. planar scene) and provided point correspondences would
* generate an inaccurate fundamental matrix.
* This implementation uses Levenberg-Marquardt algorithm for a fast cost
* optimization and uses inhomogeneous points for epipole refinement,
* which makes it unsuitable for cases when epipoles are at or near infinity
* (i.e. pure lateral translations).
*/
public class InhomogeneousRightEpipoleRefiner extends RightEpipoleRefiner {
/**
* Maximum allowed number of refinement iterations.
*/
public static final int MAX_ITERS = 10;
/**
* Constructor.
*/
public InhomogeneousRightEpipoleRefiner() {
}
/**
* Constructor.
*
* @param initialEpipoleEstimation initial right epipole estimation to be
* set and refined.
* @param keepCovariance true if covariance of estimation must be kept after
* refinement, false otherwise.
* @param inliers set indicating which of the provided matches are inliers.
* @param residuals residuals for matched samples.
* @param numInliers number of inliers on initial estimation.
* @param samples1 1st set of paired samples.
* @param samples2 2nd set of paired samples.
* @param refinementStandardDeviation standard deviation used for
* Levenberg-Marquardt fitting.
* @param homography homography relating samples in two views, which is
* used to generate a fundamental matrix and its corresponding
* epipolar geometry.
*/
public InhomogeneousRightEpipoleRefiner(
final Point2D initialEpipoleEstimation, final boolean keepCovariance, final BitSet inliers,
final double[] residuals, final int numInliers, final List<Point2D> samples1, final List<Point2D> samples2,
final double refinementStandardDeviation, final Transformation2D homography) {
super(initialEpipoleEstimation, keepCovariance, inliers, residuals, numInliers, samples1, samples2,
refinementStandardDeviation, homography);
}
/**
* Constructor.
*
* @param initialEpipoleEstimation initial right epipole estimation to be
* set and refined.
* @param keepCovariance true if covariance of estimation must be kept after
* refinement, false otherwise.
* @param inliersData inlier data, typically obtained from a robust
* estimator.
* @param samples1 1st set of paired samples.
* @param samples2 2nd set of paired samples.
* @param refinementStandardDeviation standard deviation used for
* Levenberg-Marquardt fitting.
* @param homography homography relating samples in two views, which is used
* to generate a fundamental matrix and its corresponding
* epipolar geometry.
*/
public InhomogeneousRightEpipoleRefiner(
final Point2D initialEpipoleEstimation, final boolean keepCovariance, final InliersData inliersData,
final List<Point2D> samples1, final List<Point2D> samples2, final double refinementStandardDeviation,
final Transformation2D homography) {
super(initialEpipoleEstimation, keepCovariance, inliersData, samples1, samples2, refinementStandardDeviation,
homography);
}
/**
* Refines provided initial right epipole estimation.
* This method always sets a value into provided result instance regardless
* of the fact that error has actually improved in LMSE terms or not.
*
* @param result instance where refined estimation will be stored.
* @return true if result improves (decreases) in LMSE terms respect to
* initial estimation, false if no improvement has been achieved.
* @throws NotReadyException if not enough input data has been provided.
* @throws LockedException if estimator is locked because refinement is
* already in progress.
* @throws RefinerException if refinement fails for some reason (e.g. unable
* to converge to a result).
*/
@Override
public boolean refine(final Point2D result) throws NotReadyException, LockedException, RefinerException {
if (isLocked()) {
throw new LockedException();
}
if (!isReady()) {
throw new NotReadyException();
}
locked = true;
if (listener != null) {
listener.onRefineStart(this, initialEstimation);
}
try {
final var epipole = new InhomogeneousPoint2D(initialEstimation);
var errorDecreased = false;
boolean iterErrorDecreased;
var numIter = 0;
do {
final var fundamentalMatrix = new FundamentalMatrix();
computeFundamentalMatrix(homography, epipole, fundamentalMatrix);
final var initialTotalResidual = totalResidual(fundamentalMatrix);
final var initParams = epipole.asArray();
// output values to be fitted/optimized will contain residuals
final var y = new double[numInliers];
// input values will contain 2 points to compute residuals
final var nDims = 2 * Point2D.POINT2D_INHOMOGENEOUS_COORDINATES_LENGTH;
final var x = new Matrix(numInliers, nDims);
final var nSamples = inliers.length();
var pos = 0;
Point2D leftPoint;
Point2D rightPoint;
for (var i = 0; i < nSamples; i++) {
if (inliers.get(i)) {
// sample is inlier
leftPoint = samples1.get(i);
rightPoint = samples2.get(i);
leftPoint.normalize();
rightPoint.normalize();
x.setElementAt(pos, 0, leftPoint.getInhomX());
x.setElementAt(pos, 1, leftPoint.getInhomY());
x.setElementAt(pos, 2, rightPoint.getInhomX());
x.setElementAt(pos, 3, rightPoint.getInhomY());
y[pos] = residuals[i];
pos++;
}
}
final var evaluator = new LevenbergMarquardtMultiDimensionFunctionEvaluator() {
private final Point2D leftPoint = Point2D.create(CoordinatesType.INHOMOGENEOUS_COORDINATES);
private final Point2D rightPoint = Point2D.create(CoordinatesType.INHOMOGENEOUS_COORDINATES);
private final InhomogeneousPoint2D epipole = new InhomogeneousPoint2D();
private final FundamentalMatrix fundMatrix = new FundamentalMatrix();
@SuppressWarnings("DuplicatedCode")
private final GradientEstimator gradientEstimator = new GradientEstimator(point -> {
try {
epipole.setCoordinates(point);
computeFundamentalMatrix(homography, epipole, fundMatrix);
return residual(fundMatrix, leftPoint, rightPoint);
} catch (final InvalidFundamentalMatrixException e) {
throw new EvaluationException(e);
}
});
@Override
public int getNumberOfDimensions() {
return nDims;
}
@Override
public double[] createInitialParametersArray() {
return initParams;
}
@Override
public double evaluate(
final int i, final double[] point, final double[] params, final double[] derivatives)
throws EvaluationException {
try {
leftPoint.setInhomogeneousCoordinates(point[0], point[1]);
rightPoint.setInhomogeneousCoordinates(point[2], point[3]);
epipole.setCoordinates(params);
computeFundamentalMatrix(homography, epipole, fundMatrix);
final var y = residual(fundMatrix, leftPoint, rightPoint);
gradientEstimator.gradient(params, derivatives);
return y;
} catch (final InvalidFundamentalMatrixException e) {
throw new EvaluationException(e);
}
}
};
final var fitter = new LevenbergMarquardtMultiDimensionFitter(evaluator, x, y,
getRefinementStandardDeviation());
fitter.fit();
// obtain estimated params
final var params = fitter.getA();
// update initial epipole for next iteration
epipole.setInhomogeneousCoordinates(params[0], params[1]);
computeFundamentalMatrix(homography, epipole, fundamentalMatrix);
final var finalTotalResidual = totalResidual(fundamentalMatrix);
iterErrorDecreased = finalTotalResidual < initialTotalResidual;
if (iterErrorDecreased || !errorDecreased) {
errorDecreased = true;
// update final result
result.setInhomogeneousCoordinates(params[0], params[1]);
if (keepCovariance) {
// keep covariance
covariance = fitter.getCovar();
}
}
numIter++;
} while (iterErrorDecreased && numIter < MAX_ITERS);
if (listener != null) {
listener.onRefineEnd(this, initialEstimation, result, errorDecreased);
}
return errorDecreased;
} catch (final Exception e) {
throw new RefinerException(e);
} finally {
locked = false;
}
}
}