View Javadoc
1   /*
2    * Copyright (C) 2017 Alberto Irurueta Carro (alberto@irurueta.com)
3    *
4    * Licensed under the Apache License, Version 2.0 (the "License");
5    * you may not use this file except in compliance with the License.
6    * You may obtain a copy of the License at
7    *
8    *         http://www.apache.org/licenses/LICENSE-2.0
9    *
10   * Unless required by applicable law or agreed to in writing, software
11   * distributed under the License is distributed on an "AS IS" BASIS,
12   * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13   * See the License for the specific language governing permissions and
14   * limitations under the License.
15   */
16  package com.irurueta.geometry.refiners;
17  
18  import com.irurueta.algebra.Matrix;
19  import com.irurueta.geometry.InhomogeneousPoint2D;
20  import com.irurueta.geometry.Line2D;
21  import com.irurueta.geometry.estimators.LockedException;
22  import com.irurueta.geometry.estimators.NotReadyException;
23  import com.irurueta.numerical.EvaluationException;
24  import com.irurueta.numerical.GradientEstimator;
25  import com.irurueta.numerical.fitting.LevenbergMarquardtMultiDimensionFitter;
26  import com.irurueta.numerical.fitting.LevenbergMarquardtMultiDimensionFunctionEvaluator;
27  import com.irurueta.numerical.robust.InliersData;
28  
29  import java.util.BitSet;
30  import java.util.List;
31  
32  /**
33   * Refines an inhomogeneous 2D point by taking into account an initial
34   * estimation, inlier samples and their residuals.
35   * This class can be used to find a solution that minimizes error of inliers in
36   * LMSE terms.
37   * Typically, a refiner is used by a robust estimator, however it can also be
38   * useful in some other situations.
39   */
40  @SuppressWarnings("DuplicatedCode")
41  public class InhomogeneousPoint2DRefiner extends Point2DRefiner<InhomogeneousPoint2D> {
42  
43      /**
44       * Constructor.
45       */
46      public InhomogeneousPoint2DRefiner() {
47      }
48  
49      /**
50       * Constructor.
51       *
52       * @param initialEstimation           initial estimation to be set.
53       * @param keepCovariance              true if covariance of estimation must be kept after
54       *                                    refinement, false otherwise.
55       * @param inliers                     set indicating which of the provided matches are inliers.
56       * @param residuals                   residuals for matched samples.
57       * @param numInliers                  number of inliers on initial estimation.
58       * @param samples                     collection of samples.
59       * @param refinementStandardDeviation standard deviation used for
60       *                                    Levenberg-Marquardt fitting.
61       */
62      public InhomogeneousPoint2DRefiner(
63              final InhomogeneousPoint2D initialEstimation, final boolean keepCovariance, final BitSet inliers,
64              final double[] residuals, final int numInliers, final List<Line2D> samples,
65              final double refinementStandardDeviation) {
66          super(initialEstimation, keepCovariance, inliers, residuals, numInliers, samples, refinementStandardDeviation);
67      }
68  
69      /**
70       * Constructor.
71       *
72       * @param initialEstimation           initial estimation to be set.
73       * @param keepCovariance              true if covariance of estimation must be kept after
74       *                                    refinement, false otherwise.
75       * @param inliersData                 inlier data, typically obtained from a robust
76       *                                    estimator.
77       * @param samples                     collection of samples.
78       * @param refinementStandardDeviation standard deviation used for
79       *                                    Levenberg-Marquardt fitting.
80       */
81      public InhomogeneousPoint2DRefiner(
82              final InhomogeneousPoint2D initialEstimation, final boolean keepCovariance, final InliersData inliersData,
83              final List<Line2D> samples, final double refinementStandardDeviation) {
84          super(initialEstimation, keepCovariance, inliersData, samples, refinementStandardDeviation);
85      }
86  
87      /**
88       * Refines provided initial estimation.
89       *
90       * @return refined estimation.
91       * @throws NotReadyException if not enough input data has been provided.
92       * @throws LockedException   if estimator is locked because refinement is
93       *                           already in progress.
94       * @throws RefinerException  if refinement fails for some reason (e.g. unable
95       *                           to converge to a result).
96       */
97      @Override
98      public InhomogeneousPoint2D refine() throws NotReadyException, LockedException, RefinerException {
99          final var result = new InhomogeneousPoint2D();
100         refine(result);
101         return result;
102     }
103 
104     /**
105      * Refines provided initial estimation.
106      * This method always sets a value into provided result instance regardless
107      * of the fact that error has actually improved in LMSE terms or not.
108      *
109      * @param result instance where refined estimation will be stored.
110      * @return true if result improved (decreases) in LMSE terms respect to
111      * initial estimation, false if no improvement has been achieved.
112      * @throws NotReadyException if not enough input data has been provided.
113      * @throws LockedException   if estimator is locked because refinement is
114      *                           already in progress.
115      * @throws RefinerException  if refinement fails for some reason (e.g. unable
116      *                           to converge to a result).
117      */
118     @Override
119     public boolean refine(final InhomogeneousPoint2D result) throws NotReadyException, LockedException,
120             RefinerException {
121         if (isLocked()) {
122             throw new LockedException();
123         }
124         if (!isReady()) {
125             throw new NotReadyException();
126         }
127 
128         locked = true;
129 
130         if (listener != null) {
131             listener.onRefineStart(this, initialEstimation);
132         }
133 
134         final var initialTotalResidual = totalResidual(initialEstimation);
135 
136         try {
137             final var initParams = initialEstimation.asArray();
138 
139             // output values to be fitted/optimized will contain residuals
140             final var y = new double[numInliers];
141             // input values will contain lines to compute residuals
142             final var nDims = Line2D.LINE_NUMBER_PARAMS;
143             final var x = new Matrix(numInliers, nDims);
144             final var nSamples = inliers.length();
145             var pos = 0;
146             for (var i = 0; i < nSamples; i++) {
147                 if (inliers.get(i)) {
148                     // sample is inlier
149                     final var line = samples.get(i);
150                     line.normalize();
151                     x.setElementAt(pos, 0, line.getA());
152                     x.setElementAt(pos, 1, line.getB());
153                     x.setElementAt(pos, 2, line.getC());
154 
155                     y[pos] = residuals[i];
156                     pos++;
157                 }
158             }
159 
160             final var evaluator = new LevenbergMarquardtMultiDimensionFunctionEvaluator() {
161 
162                 private final Line2D line = new Line2D();
163 
164                 private final InhomogeneousPoint2D point = new InhomogeneousPoint2D();
165 
166                 private final GradientEstimator gradientEstimator = new GradientEstimator(p -> {
167                     this.point.setCoordinates(p);
168                     return residual(this.point, line);
169                 });
170 
171                 @Override
172                 public int getNumberOfDimensions() {
173                     return nDims;
174                 }
175 
176                 @Override
177                 public double[] createInitialParametersArray() {
178                     return initParams;
179                 }
180 
181                 @Override
182                 public double evaluate(final int i, final double[] point, final double[] params,
183                                        final double[] derivatives) throws EvaluationException {
184                     // point contains, a,b,c values for line
185                     line.setParameters(point);
186 
187                     // param contains coordinates of point
188                     this.point.setCoordinates(params);
189 
190                     final var y = residual(this.point, line);
191                     gradientEstimator.gradient(params, derivatives);
192 
193                     return y;
194                 }
195             };
196 
197             final var fitter = new LevenbergMarquardtMultiDimensionFitter(evaluator, x, y,
198                     getRefinementStandardDeviation());
199 
200             fitter.fit();
201 
202             // obtain estimated params
203             final var params = fitter.getA();
204 
205             // update point
206             result.setCoordinates(params);
207 
208             if (keepCovariance) {
209                 // keep covariance
210                 covariance = fitter.getCovar();
211             }
212 
213             final var finalTotalResidual = totalResidual(result);
214             final var errorDecreased = finalTotalResidual < initialTotalResidual;
215 
216             if (listener != null) {
217                 listener.onRefineEnd(this, initialEstimation, result, errorDecreased);
218             }
219 
220             return errorDecreased;
221         } catch (final Exception e) {
222             throw new RefinerException(e);
223         } finally {
224             locked = false;
225         }
226     }
227 }