Point Cloud Library (PCL)  1.13.1-dev
ia_kfpcs.hpp
1 /*
2  * Software License Agreement (BSD License)
3  *
4  * Point Cloud Library (PCL) - www.pointclouds.org
5  * Copyright (c) 2014-, Open Perception, Inc.
6  *
7  * All rights reserved
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions are met
11  *
12  * * Redistributions of source code must retain the above copyright
13  * notice, this list of conditions and the following disclaimer.
14  * * Redistributions in binary form must reproduce the above
15  * copyright notice, this list of conditions and the following
16  * disclaimer in the documentation and/or other materials provided
17  * with the distribution.
18  * * Neither the name of the copyright holder(s) nor the names of its
19  * contributors may be used to endorse or promote products derived
20  * from this software without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
26  * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
28  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
30  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
32  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33  * POSSIBILITY OF SUCH DAMAGE.
34  *
35  */
36 
37 #ifndef PCL_REGISTRATION_IMPL_IA_KFPCS_H_
38 #define PCL_REGISTRATION_IMPL_IA_KFPCS_H_
39 
40 #include <limits>
41 
42 namespace pcl {
43 
44 namespace registration {
45 
46 template <typename PointSource, typename PointTarget, typename NormalT, typename Scalar>
49 : lower_trl_boundary_(-1.f)
50 , upper_trl_boundary_(-1.f)
51 , lambda_(0.5f)
52 , use_trl_score_(false)
53 , indices_validation_(new pcl::Indices)
54 {
55  reg_name_ = "pcl::registration::KFPCSInitialAlignment";
56 }
57 
58 template <typename PointSource, typename PointTarget, typename NormalT, typename Scalar>
59 bool
61 {
62  // due to sparse keypoint cloud, do not normalize delta with estimated point density
63  if (normalize_delta_) {
64  PCL_WARN("[%s::initCompute] Delta should be set according to keypoint precision! "
65  "Normalization according to point cloud density is ignored.\n",
66  reg_name_.c_str());
67  normalize_delta_ = false;
68  }
69 
70  // initialize as in fpcs
72  initCompute();
73 
74  // set the threshold values with respect to keypoint characteristics
75  max_pair_diff_ = delta_ * 1.414f; // diff between 2 points of delta_ accuracy
76  coincidation_limit_ = delta_ * 2.828f; // diff between diff of 2 points
77  max_edge_diff_ =
78  delta_ *
79  3.f; // diff between 2 points + some inaccuracy due to quadruple orientation
80  max_mse_ =
81  powf(delta_ * 4.f, 2.f); // diff between 2 points + some registration inaccuracy
82  max_inlier_dist_sqr_ =
83  powf(delta_ * 8.f,
84  2.f); // set rel. high, because MSAC is used (residual based score function)
85 
86  // check use of translation costs and calculate upper boundary if not set by user
87  if (upper_trl_boundary_ < 0)
88  upper_trl_boundary_ = diameter_ * (1.f - approx_overlap_) * 0.5f;
89 
90  if (!(lower_trl_boundary_ < 0) && upper_trl_boundary_ > lower_trl_boundary_)
91  use_trl_score_ = true;
92  else
93  lambda_ = 0.f;
94 
95  // generate a subset of indices of size ransac_iterations_ on which to evaluate
96  // candidates on
97  std::size_t nr_indices = indices_->size();
98  if (nr_indices < static_cast<std::size_t>(ransac_iterations_))
99  indices_validation_ = indices_;
100  else
101  for (int i = 0; i < ransac_iterations_; i++)
102  indices_validation_->push_back((*indices_)[rand() % nr_indices]);
103 
104  return (true);
105 }
106 
107 template <typename PointSource, typename PointTarget, typename NormalT, typename Scalar>
108 void
110  const pcl::Indices& base_indices,
111  std::vector<pcl::Indices>& matches,
112  MatchingCandidates& candidates)
113 {
114  candidates.clear();
115 
116  // loop over all Candidate matches
117  for (auto& match : matches) {
118  Eigen::Matrix4f transformation_temp;
119  pcl::Correspondences correspondences_temp;
120  float fitness_score =
121  std::numeric_limits<float>::max(); // reset to std::numeric_limits<float>::max()
122  // to accept all candidates and not only best
123 
124  // determine corresondences between base and match according to their distance to
125  // centroid
126  linkMatchWithBase(base_indices, match, correspondences_temp);
127 
128  // check match based on residuals of the corresponding points after transformation
129  if (validateMatch(base_indices, match, correspondences_temp, transformation_temp) <
130  0)
131  continue;
132 
133  // check resulting transformation using a sub sample of the source point cloud
134  // all candidates are stored and later sorted according to their fitness score
135  validateTransformation(transformation_temp, fitness_score);
136 
137  // store all valid match as well as associated score and transformation
138  candidates.push_back(
139  MatchingCandidate(fitness_score, correspondences_temp, transformation_temp));
140  }
141 }
142 
143 template <typename PointSource, typename PointTarget, typename NormalT, typename Scalar>
144 int
146  validateTransformation(Eigen::Matrix4f& transformation, float& fitness_score)
147 {
148  // transform sub sampled source cloud
149  PointCloudSource source_transformed;
151  *input_, *indices_validation_, source_transformed, transformation);
152 
153  const std::size_t nr_points = source_transformed.size();
154  float score_a = 0.f, score_b = 0.f;
155 
156  // residual costs based on mse
157  pcl::Indices ids;
158  std::vector<float> dists_sqr;
159  for (const auto& source : source_transformed) {
160  // search for nearest point using kd tree search
161  tree_->nearestKSearch(source, 1, ids, dists_sqr);
162  score_a += (dists_sqr[0] < max_inlier_dist_sqr_ ? dists_sqr[0]
163  : max_inlier_dist_sqr_); // MSAC
164  }
165 
166  score_a /= (max_inlier_dist_sqr_ * nr_points); // MSAC
167  // score_a = 1.f - (1.f - score_a) / (1.f - approx_overlap_); // make score relative
168  // to estimated overlap
169 
170  // translation score (solutions with small translation are down-voted)
171  float scale = 1.f;
172  if (use_trl_score_) {
173  float trl = transformation.rightCols<1>().head(3).norm();
174  float trl_ratio =
175  (trl - lower_trl_boundary_) / (upper_trl_boundary_ - lower_trl_boundary_);
176 
177  score_b =
178  (trl_ratio < 0.f ? 1.f
179  : (trl_ratio > 1.f ? 0.f
180  : 0.5f * sin(M_PI * trl_ratio + M_PI_2) +
181  0.5f)); // sinusoidal costs
182  scale += lambda_;
183  }
184 
185  // calculate the fitness and return unsuccessful if smaller than previous ones
186  float fitness_score_temp = (score_a + lambda_ * score_b) / scale;
187  if (fitness_score_temp > fitness_score)
188  return (-1);
189 
190  fitness_score = fitness_score_temp;
191  return (0);
192 }
193 
194 template <typename PointSource, typename PointTarget, typename NormalT, typename Scalar>
195 void
197  const std::vector<MatchingCandidates>& candidates)
198 {
199  // reorganize candidates into single vector
200  std::size_t total_size = 0;
201  for (const auto& candidate : candidates)
202  total_size += candidate.size();
203 
204  candidates_.clear();
205  candidates_.reserve(total_size);
206 
207  for (const auto& candidate : candidates)
208  for (const auto& match : candidate)
209  candidates_.push_back(match);
210 
211  // sort according to score value
212  std::sort(candidates_.begin(), candidates_.end(), by_score());
213 
214  // return here if no score was valid, i.e. all scores are
215  // std::numeric_limits<float>::max()
216  if (candidates_[0].fitness_score == std::numeric_limits<float>::max()) {
217  converged_ = false;
218  return;
219  }
220 
221  // save best candidate as output result
222  // note, all other candidates are accessible via getNBestCandidates () and
223  // getTBestCandidates ()
224  fitness_score_ = candidates_[0].fitness_score;
225  final_transformation_ = candidates_[0].transformation;
226  *correspondences_ = candidates_[0].correspondences;
227 
228  // here we define convergence if resulting score is above threshold
229  converged_ = fitness_score_ < score_threshold_;
230 }
231 
232 template <typename PointSource, typename PointTarget, typename NormalT, typename Scalar>
233 void
235  int n, float min_angle3d, float min_translation3d, MatchingCandidates& candidates)
236 {
237  candidates.clear();
238 
239  // loop over all candidates starting from the best one
240  for (const auto& candidate : candidates_) {
241  // stop if current candidate has no valid score
242  if (candidate.fitness_score == std::numeric_limits<float>::max())
243  return;
244 
245  // check if current candidate is a unique one compared to previous using the
246  // min_diff threshold
247  bool unique = true;
248  for (const auto& c2 : candidates) {
249  Eigen::Matrix4f diff =
250  candidate.transformation.colPivHouseholderQr().solve(c2.transformation);
251  const float angle3d = Eigen::AngleAxisf(diff.block<3, 3>(0, 0)).angle();
252  const float translation3d = diff.block<3, 1>(0, 3).norm();
253  unique = angle3d > min_angle3d && translation3d > min_translation3d;
254  if (!unique) {
255  break;
256  }
257  }
258 
259  // add candidate to best candidates
260  if (unique)
261  candidates.push_back(candidate);
262 
263  // stop if n candidates are reached
264  if (candidates.size() == n)
265  return;
266  }
267 }
268 
269 template <typename PointSource, typename PointTarget, typename NormalT, typename Scalar>
270 void
272  float t, float min_angle3d, float min_translation3d, MatchingCandidates& candidates)
273 {
274  candidates.clear();
275 
276  // loop over all candidates starting from the best one
277  for (const auto& candidate : candidates_) {
278  // stop if current candidate has score below threshold
279  if (candidate.fitness_score > t)
280  return;
281 
282  // check if current candidate is a unique one compared to previous using the
283  // min_diff threshold
284  bool unique = true;
285  for (const auto& c2 : candidates) {
286  Eigen::Matrix4f diff =
287  candidate.transformation.colPivHouseholderQr().solve(c2.transformation);
288  const float angle3d = Eigen::AngleAxisf(diff.block<3, 3>(0, 0)).angle();
289  const float translation3d = diff.block<3, 1>(0, 3).norm();
290  unique = angle3d > min_angle3d && translation3d > min_translation3d;
291  if (!unique) {
292  break;
293  }
294  }
295 
296  // add candidate to best candidates
297  if (unique)
298  candidates.push_back(candidate);
299  }
300 }
301 
302 } // namespace registration
303 } // namespace pcl
304 
305 #endif // PCL_REGISTRATION_IMPL_IA_KFPCS_H_
std::size_t size() const
Definition: point_cloud.h:443
std::string reg_name_
The registration method name.
Definition: registration.h:560
virtual bool initCompute()
Internal computation initialization.
Definition: ia_fpcs.hpp:242
void getTBestCandidates(float t, float min_angle3d, float min_translation3d, MatchingCandidates &candidates)
Get all unique candidate matches with fitness scores above a threshold t.
Definition: ia_kfpcs.hpp:271
void finalCompute(const std::vector< MatchingCandidates > &candidates) override
Final computation of best match out of vector of matches.
Definition: ia_kfpcs.hpp:196
void handleMatches(const pcl::Indices &base_indices, std::vector< pcl::Indices > &matches, MatchingCandidates &candidates) override
Method to handle current candidate matches.
Definition: ia_kfpcs.hpp:109
void getNBestCandidates(int n, float min_angle3d, float min_translation3d, MatchingCandidates &candidates)
Get the N best unique candidate matches according to their fitness score.
Definition: ia_kfpcs.hpp:234
int validateTransformation(Eigen::Matrix4f &transformation, float &fitness_score) override
Validate the transformation by calculating the score value after transforming the input source cloud.
Definition: ia_kfpcs.hpp:146
bool initCompute() override
Internal computation initialization.
Definition: ia_kfpcs.hpp:60
void transformPointCloud(const pcl::PointCloud< PointT > &cloud_in, pcl::PointCloud< PointT > &cloud_out, const Eigen::Matrix< Scalar, 4, 4 > &transform, bool copy_all_fields)
Apply a rigid transform defined by a 4x4 matrix.
Definition: transforms.hpp:221
std::vector< MatchingCandidate, Eigen::aligned_allocator< MatchingCandidate > > MatchingCandidates
std::vector< pcl::Correspondence, Eigen::aligned_allocator< pcl::Correspondence > > Correspondences
IndicesAllocator<> Indices
Type used for indices in PCL.
Definition: types.h:133
#define M_PI_2
Definition: pcl_macros.h:202
#define M_PI
Definition: pcl_macros.h:201
Container for matching candidate consisting of.
Sorting of candidates based on fitness score value.