View Javadoc
1   /*
2    * Copyright (C) 2015 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.estimators;
17  
18  import com.irurueta.geometry.CoincidentPointsException;
19  import com.irurueta.geometry.Conic;
20  import com.irurueta.geometry.Point2D;
21  import com.irurueta.numerical.robust.LMedSRobustEstimator;
22  import com.irurueta.numerical.robust.LMedSRobustEstimatorListener;
23  import com.irurueta.numerical.robust.RobustEstimator;
24  import com.irurueta.numerical.robust.RobustEstimatorException;
25  import com.irurueta.numerical.robust.RobustEstimatorMethod;
26  
27  import java.util.List;
28  
29  /**
30   * Finds the best conic for provided collection of 2D points using LMedS
31   * algorithm.
32   */
33  @SuppressWarnings("DuplicatedCode")
34  public class LMedSConicRobustEstimator extends ConicRobustEstimator {
35  
36      /**
37       * Default value to be used for stop threshold. Stop threshold can be used
38       * to keep the algorithm iterating in case that best estimated threshold
39       * using median of residuals is not small enough. Once a solution is found
40       * that generates a threshold below this value, the algorithm will stop.
41       * The stop threshold can be used to prevent the LMedS algorithm iterating
42       * too many times in cases where samples have a very similar accuracy.
43       * For instance, in cases where proportion of outliers is very small (close
44       * to 0%), and samples are very accurate (i.e. 1e-6), the algorithm would
45       * iterate for a long time trying to find the best solution when indeed
46       * there is no need to do that if a reasonable threshold has already been
47       * reached.
48       * Because of this behaviour the stop threshold can be set to a value much
49       * lower than the one typically used in RANSAC, and yet the algorithm could
50       * still produce even smaller thresholds in estimated results.
51       */
52      public static final double DEFAULT_STOP_THRESHOLD = 1e-9;
53  
54      /**
55       * Minimum allowed stop threshold value.
56       */
57      public static final double MIN_STOP_THRESHOLD = 0.0;
58  
59      /**
60       * Threshold to be used to keep the algorithm iterating in case that best
61       * estimated threshold using median of residuals is not small enough. Once
62       * a solution is found that generates a threshold below this value, the
63       * algorithm will stop.
64       * The stop threshold can be used to prevent the LMedS algorithm iterating
65       * too many times in cases where samples have a very similar accuracy.
66       * For instance, in cases where proportion of outliers is very small (close
67       * to 0%), and samples are very accurate (i.e. 1e-6), the algorithm would
68       * iterate for a long time trying to find the best solution when indeed
69       * there is no need to do that if a reasonable threshold has already been
70       * reached.
71       * Because of this behaviour the stop threshold can be set to a value much
72       * lower than the one typically used in RANSAC, and yet the algorithm could
73       * still produce even smaller thresholds in estimated results.
74       */
75      private double stopThreshold;
76  
77      /**
78       * Constructor.
79       */
80      public LMedSConicRobustEstimator() {
81          super();
82          stopThreshold = DEFAULT_STOP_THRESHOLD;
83      }
84  
85      /**
86       * Constructor with points.
87       *
88       * @param points 2D points to estimate a conic.
89       * @throws IllegalArgumentException if provided list of points don't have
90       *                                  a size greater or equal than MINIMUM_SIZE.
91       */
92      public LMedSConicRobustEstimator(final List<Point2D> points) {
93          super(points);
94          stopThreshold = DEFAULT_STOP_THRESHOLD;
95      }
96  
97      /**
98       * Constructor.
99       *
100      * @param listener listener to be notified of events such as when estimation
101      *                 starts, ends or its progress significantly changes.
102      */
103     public LMedSConicRobustEstimator(final ConicRobustEstimatorListener listener) {
104         super(listener);
105         stopThreshold = DEFAULT_STOP_THRESHOLD;
106     }
107 
108     /**
109      * Constructor.
110      *
111      * @param listener listener to be notified of events such as when estimation
112      *                 starts, ends or its progress significantly changes.
113      * @param points   2D points to estimate a conic.
114      * @throws IllegalArgumentException if provided list of points don't have a
115      *                                  size greater or equal than MINIMUM_SIZE.
116      */
117     public LMedSConicRobustEstimator(final ConicRobustEstimatorListener listener, final List<Point2D> points) {
118         super(listener, points);
119         stopThreshold = DEFAULT_STOP_THRESHOLD;
120     }
121 
122     /**
123      * Returns threshold to be used to keep the algorithm iterating in case that
124      * best estimated threshold using median of residuals is not small enough.
125      * Once a solution is found that generates a threshold below this value, the
126      * algorithm will stop.
127      * The stop threshold can be used to prevent the LMedS algorithm iterating
128      * too many times in cases where samples have a very similar accuracy.
129      * For instance, in cases where proportion of outliers is very small (close
130      * to 0%), and samples are very accurate (i.e. 1e-6), the algorithm would
131      * iterate for a long time trying to find the best solution when indeed
132      * there is no need to do that if a reasonable threshold has already been
133      * reached.
134      * Because of this behaviour the stop threshold can be set to a value much
135      * lower than the one typically used in RANSAC, and yet the algorithm could
136      * still produce even smaller thresholds in estimated results.
137      *
138      * @return stop threshold to stop the algorithm prematurely when a certain
139      * accuracy has been reached.
140      */
141     public double getStopThreshold() {
142         return stopThreshold;
143     }
144 
145     /**
146      * Sets threshold to be used to keep the algorithm iterating in case that
147      * best estimated threshold using median of residuals is not small enough.
148      * Once a solution is found that generates a threshold below this value, the
149      * algorithm will stop.
150      * The stop threshold can be used to prevent the LMedS algorithm iterating
151      * too many times in cases where samples have a very similar accuracy.
152      * For instance, in cases where proportion of outliers is very small (close
153      * to 0%), and samples are very accurate (i.e. 1e-6), the algorithm would
154      * iterate for a long time trying to find the best solution when indeed
155      * there is no need to do that if a reasonable threshold has already been
156      * reached.
157      * Because of this behaviour the stop threshold can be set to a value much
158      * lower than the one typically used in RANSAC, and yet the algorithm could
159      * still produce even smaller thresholds in estimated results.
160      *
161      * @param stopThreshold stop threshold to stop the algorithm prematurely
162      *                      when a certain accuracy has been reached.
163      * @throws IllegalArgumentException if provided value is zero or negative.
164      * @throws LockedException          if robust estimator is locked because an
165      *                                  estimation is already in progress.
166      */
167     public void setStopThreshold(final double stopThreshold) throws LockedException {
168         if (isLocked()) {
169             throw new LockedException();
170         }
171         if (stopThreshold <= MIN_STOP_THRESHOLD) {
172             throw new IllegalArgumentException();
173         }
174 
175         this.stopThreshold = stopThreshold;
176     }
177 
178     /**
179      * Estimates a conic using a robust estimator and the best set of 2D points
180      * that fit into the locus of the estimated conic found using the robust
181      * estimator.
182      *
183      * @return a conic.
184      * @throws LockedException          if robust estimator is locked because an
185      *                                  estimation is already in progress.
186      * @throws NotReadyException        if provided input data is not enough to start
187      *                                  the estimation.
188      * @throws RobustEstimatorException if estimation fails for any reason
189      *                                  (i.e. numerical instability, no solution available, etc).
190      */
191     @Override
192     public Conic estimate() throws LockedException, NotReadyException, RobustEstimatorException {
193         if (isLocked()) {
194             throw new LockedException();
195         }
196         if (!isReady()) {
197             throw new NotReadyException();
198         }
199 
200         final var innerEstimator = new LMedSRobustEstimator<>(new LMedSRobustEstimatorListener<Conic>() {
201 
202             @Override
203             public int getTotalSamples() {
204                 return points.size();
205             }
206 
207             @Override
208             public int getSubsetSize() {
209                 return ConicRobustEstimator.MINIMUM_SIZE;
210             }
211 
212             @Override
213             public void estimatePreliminarSolutions(final int[] samplesIndices, final List<Conic> solutions) {
214                 final var point1 = points.get(samplesIndices[0]);
215                 final var point2 = points.get(samplesIndices[1]);
216                 final var point3 = points.get(samplesIndices[2]);
217                 final var point4 = points.get(samplesIndices[3]);
218                 final var point5 = points.get(samplesIndices[4]);
219 
220                 try {
221                     final var conic = new Conic(point1, point2, point3, point4, point5);
222                     solutions.add(conic);
223                 } catch (final CoincidentPointsException e) {
224                     // if points are coincident, no solution is added
225                 }
226             }
227 
228             @Override
229             public double computeResidual(final Conic currentEstimation, final int i) {
230                 return residual(currentEstimation, points.get(i));
231             }
232 
233             @Override
234             public boolean isReady() {
235                 return LMedSConicRobustEstimator.this.isReady();
236             }
237 
238             @Override
239             public void onEstimateStart(final RobustEstimator<Conic> estimator) {
240                 if (listener != null) {
241                     listener.onEstimateStart(LMedSConicRobustEstimator.this);
242                 }
243             }
244 
245             @Override
246             public void onEstimateEnd(final RobustEstimator<Conic> estimator) {
247                 if (listener != null) {
248                     listener.onEstimateEnd(LMedSConicRobustEstimator.this);
249                 }
250             }
251 
252             @Override
253             public void onEstimateNextIteration(final RobustEstimator<Conic> estimator, final int iteration) {
254                 if (listener != null) {
255                     listener.onEstimateNextIteration(LMedSConicRobustEstimator.this, iteration);
256                 }
257             }
258 
259             @Override
260             public void onEstimateProgressChange(final RobustEstimator<Conic> estimator, final float progress) {
261                 if (listener != null) {
262                     listener.onEstimateProgressChange(LMedSConicRobustEstimator.this, progress);
263                 }
264             }
265         });
266 
267         try {
268             locked = true;
269             innerEstimator.setConfidence(confidence);
270             innerEstimator.setMaxIterations(maxIterations);
271             innerEstimator.setProgressDelta(progressDelta);
272             innerEstimator.setStopThreshold(stopThreshold);
273             return innerEstimator.estimate();
274         } catch (final com.irurueta.numerical.LockedException e) {
275             throw new LockedException(e);
276         } catch (final com.irurueta.numerical.NotReadyException e) {
277             throw new NotReadyException(e);
278         } finally {
279             locked = false;
280         }
281     }
282 
283     /**
284      * Returns method being used for robust estimation.
285      *
286      * @return method being used for robust estimation.
287      */
288     @Override
289     public RobustEstimatorMethod getMethod() {
290         return RobustEstimatorMethod.LMEDS;
291     }
292 }