Point Cloud Library (PCL)  1.11.1-dev
gicp.hpp
1 /*
2  * Software License Agreement (BSD License)
3  *
4  * Point Cloud Library (PCL) - www.pointclouds.org
5  * Copyright (c) 2010, Willow Garage, Inc.
6  * Copyright (c) 2012-, Open Perception, Inc.
7  *
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  *
14  * * Redistributions of source code must retain the above copyright
15  * notice, this list of conditions and the following disclaimer.
16  * * Redistributions in binary form must reproduce the above
17  * copyright notice, this list of conditions and the following
18  * disclaimer in the documentation and/or other materials provided
19  * with the distribution.
20  * * Neither the name of the copyright holder(s) nor the names of its
21  * contributors may be used to endorse or promote products derived
22  * from this software without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
25  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
26  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
27  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
28  * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
29  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
30  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
31  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
32  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
34  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35  * POSSIBILITY OF SUCH DAMAGE.
36  *
37  * $Id$
38  *
39  */
40 
41 #ifndef PCL_REGISTRATION_IMPL_GICP_HPP_
42 #define PCL_REGISTRATION_IMPL_GICP_HPP_
43 
44 #include <pcl/registration/exceptions.h>
45 
46 namespace pcl {
47 
48 template <typename PointSource, typename PointTarget>
49 template <typename PointT>
50 void
51 GeneralizedIterativeClosestPoint<PointSource, PointTarget>::computeCovariances(
52  typename pcl::PointCloud<PointT>::ConstPtr cloud,
53  const typename pcl::search::KdTree<PointT>::Ptr kdtree,
54  MatricesVector& cloud_covariances)
55 {
56  if (k_correspondences_ > int(cloud->size())) {
57  PCL_ERROR("[pcl::GeneralizedIterativeClosestPoint::computeCovariances] Number or "
58  "points in cloud (%lu) is less than k_correspondences_ (%lu)!\n",
59  cloud->size(),
60  k_correspondences_);
61  return;
62  }
63 
64  Eigen::Vector3d mean;
65  pcl::Indices nn_indecies;
66  nn_indecies.reserve(k_correspondences_);
67  std::vector<float> nn_dist_sq;
68  nn_dist_sq.reserve(k_correspondences_);
69 
70  // We should never get there but who knows
71  if (cloud_covariances.size() < cloud->size())
72  cloud_covariances.resize(cloud->size());
73 
74  MatricesVector::iterator matrices_iterator = cloud_covariances.begin();
75  for (auto points_iterator = cloud->begin(); points_iterator != cloud->end();
76  ++points_iterator, ++matrices_iterator) {
77  const PointT& query_point = *points_iterator;
78  Eigen::Matrix3d& cov = *matrices_iterator;
79  // Zero out the cov and mean
80  cov.setZero();
81  mean.setZero();
82 
83  // Search for the K nearest neighbours
84  kdtree->nearestKSearch(query_point, k_correspondences_, nn_indecies, nn_dist_sq);
85 
86  // Find the covariance matrix
87  for (int j = 0; j < k_correspondences_; j++) {
88  const PointT& pt = (*cloud)[nn_indecies[j]];
89 
90  mean[0] += pt.x;
91  mean[1] += pt.y;
92  mean[2] += pt.z;
93 
94  cov(0, 0) += pt.x * pt.x;
95 
96  cov(1, 0) += pt.y * pt.x;
97  cov(1, 1) += pt.y * pt.y;
98 
99  cov(2, 0) += pt.z * pt.x;
100  cov(2, 1) += pt.z * pt.y;
101  cov(2, 2) += pt.z * pt.z;
102  }
103 
104  mean /= static_cast<double>(k_correspondences_);
105  // Get the actual covariance
106  for (int k = 0; k < 3; k++)
107  for (int l = 0; l <= k; l++) {
108  cov(k, l) /= static_cast<double>(k_correspondences_);
109  cov(k, l) -= mean[k] * mean[l];
110  cov(l, k) = cov(k, l);
111  }
112 
113  // Compute the SVD (covariance matrix is symmetric so U = V')
114  Eigen::JacobiSVD<Eigen::Matrix3d> svd(cov, Eigen::ComputeFullU);
115  cov.setZero();
116  Eigen::Matrix3d U = svd.matrixU();
117  // Reconstitute the covariance matrix with modified singular values using the column
118  // // vectors in V.
119  for (int k = 0; k < 3; k++) {
120  Eigen::Vector3d col = U.col(k);
121  double v = 1.; // biggest 2 singular values replaced by 1
122  if (k == 2) // smallest singular value replaced by gicp_epsilon
123  v = gicp_epsilon_;
124  cov += v * col * col.transpose();
125  }
126  }
127 }
128 
129 template <typename PointSource, typename PointTarget>
130 void
131 GeneralizedIterativeClosestPoint<PointSource, PointTarget>::computeRDerivative(
132  const Vector6d& x, const Eigen::Matrix3d& R, Vector6d& g) const
133 {
134  Eigen::Matrix3d dR_dPhi;
135  Eigen::Matrix3d dR_dTheta;
136  Eigen::Matrix3d dR_dPsi;
137 
138  double phi = x[3], theta = x[4], psi = x[5];
139 
140  double cphi = std::cos(phi), sphi = sin(phi);
141  double ctheta = std::cos(theta), stheta = sin(theta);
142  double cpsi = std::cos(psi), spsi = sin(psi);
143 
144  dR_dPhi(0, 0) = 0.;
145  dR_dPhi(1, 0) = 0.;
146  dR_dPhi(2, 0) = 0.;
147 
148  dR_dPhi(0, 1) = sphi * spsi + cphi * cpsi * stheta;
149  dR_dPhi(1, 1) = -cpsi * sphi + cphi * spsi * stheta;
150  dR_dPhi(2, 1) = cphi * ctheta;
151 
152  dR_dPhi(0, 2) = cphi * spsi - cpsi * sphi * stheta;
153  dR_dPhi(1, 2) = -cphi * cpsi - sphi * spsi * stheta;
154  dR_dPhi(2, 2) = -ctheta * sphi;
155 
156  dR_dTheta(0, 0) = -cpsi * stheta;
157  dR_dTheta(1, 0) = -spsi * stheta;
158  dR_dTheta(2, 0) = -ctheta;
159 
160  dR_dTheta(0, 1) = cpsi * ctheta * sphi;
161  dR_dTheta(1, 1) = ctheta * sphi * spsi;
162  dR_dTheta(2, 1) = -sphi * stheta;
163 
164  dR_dTheta(0, 2) = cphi * cpsi * ctheta;
165  dR_dTheta(1, 2) = cphi * ctheta * spsi;
166  dR_dTheta(2, 2) = -cphi * stheta;
167 
168  dR_dPsi(0, 0) = -ctheta * spsi;
169  dR_dPsi(1, 0) = cpsi * ctheta;
170  dR_dPsi(2, 0) = 0.;
171 
172  dR_dPsi(0, 1) = -cphi * cpsi - sphi * spsi * stheta;
173  dR_dPsi(1, 1) = -cphi * spsi + cpsi * sphi * stheta;
174  dR_dPsi(2, 1) = 0.;
175 
176  dR_dPsi(0, 2) = cpsi * sphi - cphi * spsi * stheta;
177  dR_dPsi(1, 2) = sphi * spsi + cphi * cpsi * stheta;
178  dR_dPsi(2, 2) = 0.;
179 
180  g[3] = matricesInnerProd(dR_dPhi, R);
181  g[4] = matricesInnerProd(dR_dTheta, R);
182  g[5] = matricesInnerProd(dR_dPsi, R);
183 }
184 
185 template <typename PointSource, typename PointTarget>
186 void
187 GeneralizedIterativeClosestPoint<PointSource, PointTarget>::
188  estimateRigidTransformationBFGS(const PointCloudSource& cloud_src,
189  const std::vector<int>& indices_src,
190  const PointCloudTarget& cloud_tgt,
191  const std::vector<int>& indices_tgt,
192  Eigen::Matrix4f& transformation_matrix)
193 {
194  if (indices_src.size() < 4) // need at least 4 samples
195  {
196  PCL_THROW_EXCEPTION(
197  NotEnoughPointsException,
198  "[pcl::GeneralizedIterativeClosestPoint::estimateRigidTransformationBFGS] Need "
199  "at least 4 points to estimate a transform! Source and target have "
200  << indices_src.size() << " points!");
201  return;
202  }
203  // Set the initial solution
204  Vector6d x = Vector6d::Zero();
205  // translation part
206  x[0] = transformation_matrix(0, 3);
207  x[1] = transformation_matrix(1, 3);
208  x[2] = transformation_matrix(2, 3);
209  // rotation part (Z Y X euler angles convention)
210  // see: https://en.wikipedia.org/wiki/Rotation_matrix#General_rotations
211  x[3] = std::atan2(transformation_matrix(2, 1), transformation_matrix(2, 2));
212  x[4] = asin(-transformation_matrix(2, 0));
213  x[5] = std::atan2(transformation_matrix(1, 0), transformation_matrix(0, 0));
214 
215  // Set temporary pointers
216  tmp_src_ = &cloud_src;
217  tmp_tgt_ = &cloud_tgt;
218  tmp_idx_src_ = &indices_src;
219  tmp_idx_tgt_ = &indices_tgt;
220 
221  // Optimize using forward-difference approximation LM
222  OptimizationFunctorWithIndices functor(this);
223  BFGS<OptimizationFunctorWithIndices> bfgs(functor);
224  bfgs.parameters.sigma = 0.01;
225  bfgs.parameters.rho = 0.01;
226  bfgs.parameters.tau1 = 9;
227  bfgs.parameters.tau2 = 0.05;
228  bfgs.parameters.tau3 = 0.5;
229  bfgs.parameters.order = 3;
230 
231  int inner_iterations_ = 0;
232  int result = bfgs.minimizeInit(x);
233  result = BFGSSpace::Running;
234  do {
235  inner_iterations_++;
236  result = bfgs.minimizeOneStep(x);
237  if (result) {
238  break;
239  }
240  result = bfgs.testGradient();
241  } while (result == BFGSSpace::Running && inner_iterations_ < max_inner_iterations_);
242  if (result == BFGSSpace::NoProgress || result == BFGSSpace::Success ||
243  inner_iterations_ == max_inner_iterations_) {
244  PCL_DEBUG("[pcl::registration::TransformationEstimationBFGS::"
245  "estimateRigidTransformation]");
246  PCL_DEBUG("BFGS solver finished with exit code %i \n", result);
247  transformation_matrix.setIdentity();
248  applyState(transformation_matrix, x);
249  }
250  else
251  PCL_THROW_EXCEPTION(
252  SolverDidntConvergeException,
253  "[pcl::" << getClassName()
254  << "::TransformationEstimationBFGS::estimateRigidTransformation] BFGS "
255  "solver didn't converge!");
256 }
257 
258 template <typename PointSource, typename PointTarget>
259 inline double
260 GeneralizedIterativeClosestPoint<PointSource, PointTarget>::
261  OptimizationFunctorWithIndices::operator()(const Vector6d& x)
262 {
263  Eigen::Matrix4f transformation_matrix = gicp_->base_transformation_;
264  gicp_->applyState(transformation_matrix, x);
265  double f = 0;
266  int m = static_cast<int>(gicp_->tmp_idx_src_->size());
267  for (int i = 0; i < m; ++i) {
268  // The last coordinate, p_src[3] is guaranteed to be set to 1.0 in registration.hpp
269  Vector4fMapConst p_src =
270  (*gicp_->tmp_src_)[(*gicp_->tmp_idx_src_)[i]].getVector4fMap();
271  // The last coordinate, p_tgt[3] is guaranteed to be set to 1.0 in registration.hpp
272  Vector4fMapConst p_tgt =
273  (*gicp_->tmp_tgt_)[(*gicp_->tmp_idx_tgt_)[i]].getVector4fMap();
274  Eigen::Vector4f pp(transformation_matrix * p_src);
275  // Estimate the distance (cost function)
276  // The last coordinate is still guaranteed to be set to 1.0
277  Eigen::Vector3d res(pp[0] - p_tgt[0], pp[1] - p_tgt[1], pp[2] - p_tgt[2]);
278  Eigen::Vector3d temp(gicp_->mahalanobis((*gicp_->tmp_idx_src_)[i]) * res);
279  // increment= res'*temp/num_matches = temp'*M*temp/num_matches (we postpone
280  // 1/num_matches after the loop closes)
281  f += double(res.transpose() * temp);
282  }
283  return f / m;
284 }
285 
286 template <typename PointSource, typename PointTarget>
287 inline void
288 GeneralizedIterativeClosestPoint<PointSource, PointTarget>::
289  OptimizationFunctorWithIndices::df(const Vector6d& x, Vector6d& g)
290 {
291  Eigen::Matrix4f transformation_matrix = gicp_->base_transformation_;
292  gicp_->applyState(transformation_matrix, x);
293  // Zero out g
294  g.setZero();
295  // Eigen::Vector3d g_t = g.head<3> ();
296  Eigen::Matrix3d R = Eigen::Matrix3d::Zero();
297  int m = static_cast<int>(gicp_->tmp_idx_src_->size());
298  for (int i = 0; i < m; ++i) {
299  // The last coordinate, p_src[3] is guaranteed to be set to 1.0 in registration.hpp
300  Vector4fMapConst p_src =
301  (*gicp_->tmp_src_)[(*gicp_->tmp_idx_src_)[i]].getVector4fMap();
302  // The last coordinate, p_tgt[3] is guaranteed to be set to 1.0 in registration.hpp
303  Vector4fMapConst p_tgt =
304  (*gicp_->tmp_tgt_)[(*gicp_->tmp_idx_tgt_)[i]].getVector4fMap();
305 
306  Eigen::Vector4f pp(transformation_matrix * p_src);
307  // The last coordinate is still guaranteed to be set to 1.0
308  Eigen::Vector3d res(pp[0] - p_tgt[0], pp[1] - p_tgt[1], pp[2] - p_tgt[2]);
309  // temp = M*res
310  Eigen::Vector3d temp(gicp_->mahalanobis((*gicp_->tmp_idx_src_)[i]) * res);
311  // Increment translation gradient
312  // g.head<3> ()+= 2*M*res/num_matches (we postpone 2/num_matches after the loop
313  // closes)
314  g.head<3>() += temp;
315  // Increment rotation gradient
316  pp = gicp_->base_transformation_ * p_src;
317  Eigen::Vector3d p_src3(pp[0], pp[1], pp[2]);
318  R += p_src3 * temp.transpose();
319  }
320  g.head<3>() *= 2.0 / m;
321  R *= 2.0 / m;
322  gicp_->computeRDerivative(x, R, g);
323 }
324 
325 template <typename PointSource, typename PointTarget>
326 inline void
327 GeneralizedIterativeClosestPoint<PointSource, PointTarget>::
328  OptimizationFunctorWithIndices::fdf(const Vector6d& x, double& f, Vector6d& g)
329 {
330  Eigen::Matrix4f transformation_matrix = gicp_->base_transformation_;
331  gicp_->applyState(transformation_matrix, x);
332  f = 0;
333  g.setZero();
334  Eigen::Matrix3d R = Eigen::Matrix3d::Zero();
335  const int m = static_cast<int>(gicp_->tmp_idx_src_->size());
336  for (int i = 0; i < m; ++i) {
337  // The last coordinate, p_src[3] is guaranteed to be set to 1.0 in registration.hpp
338  Vector4fMapConst p_src =
339  (*gicp_->tmp_src_)[(*gicp_->tmp_idx_src_)[i]].getVector4fMap();
340  // The last coordinate, p_tgt[3] is guaranteed to be set to 1.0 in registration.hpp
341  Vector4fMapConst p_tgt =
342  (*gicp_->tmp_tgt_)[(*gicp_->tmp_idx_tgt_)[i]].getVector4fMap();
343  Eigen::Vector4f pp(transformation_matrix * p_src);
344  // The last coordinate is still guaranteed to be set to 1.0
345  Eigen::Vector3d res(pp[0] - p_tgt[0], pp[1] - p_tgt[1], pp[2] - p_tgt[2]);
346  // temp = M*res
347  Eigen::Vector3d temp(gicp_->mahalanobis((*gicp_->tmp_idx_src_)[i]) * res);
348  // Increment total error
349  f += double(res.transpose() * temp);
350  // Increment translation gradient
351  // g.head<3> ()+= 2*M*res/num_matches (we postpone 2/num_matches after the loop
352  // closes)
353  g.head<3>() += temp;
354  pp = gicp_->base_transformation_ * p_src;
355  Eigen::Vector3d p_src3(pp[0], pp[1], pp[2]);
356  // Increment rotation gradient
357  R += p_src3 * temp.transpose();
358  }
359  f /= double(m);
360  g.head<3>() *= double(2.0 / m);
361  R *= 2.0 / m;
362  gicp_->computeRDerivative(x, R, g);
363 }
364 
365 template <typename PointSource, typename PointTarget>
366 inline BFGSSpace::Status
367 GeneralizedIterativeClosestPoint<PointSource, PointTarget>::
368  OptimizationFunctorWithIndices::checkGradient(const Vector6d& g)
369 {
370  auto translation_epsilon = gicp_->translation_gradient_tolerance_;
371  auto rotation_epsilon = gicp_->rotation_gradient_tolerance_;
372 
373  if ((translation_epsilon < 0.) || (rotation_epsilon < 0.))
374  return BFGSSpace::NegativeGradientEpsilon;
375 
376  // express translation gradient as norm of translation parameters
377  auto translation_grad = g.head<3>().norm();
378 
379  // express rotation gradient as a norm of rotation parameters
380  auto rotation_grad = g.tail<3>().norm();
381 
382  if ((translation_grad < translation_epsilon) && (rotation_grad < rotation_epsilon))
383  return BFGSSpace::Success;
384 
385  return BFGSSpace::Running;
386 }
387 
388 template <typename PointSource, typename PointTarget>
389 inline void
390 GeneralizedIterativeClosestPoint<PointSource, PointTarget>::computeTransformation(
391  PointCloudSource& output, const Eigen::Matrix4f& guess)
392 {
393  pcl::IterativeClosestPoint<PointSource, PointTarget>::initComputeReciprocal();
394  // Difference between consecutive transforms
395  double delta = 0;
396  // Get the size of the target
397  const std::size_t N = indices_->size();
398  // Set the mahalanobis matrices to identity
399  mahalanobis_.resize(N, Eigen::Matrix3d::Identity());
400  // Compute target cloud covariance matrices
401  if ((!target_covariances_) || (target_covariances_->empty())) {
402  target_covariances_.reset(new MatricesVector);
403  computeCovariances<PointTarget>(target_, tree_, *target_covariances_);
404  }
405  // Compute input cloud covariance matrices
406  if ((!input_covariances_) || (input_covariances_->empty())) {
407  input_covariances_.reset(new MatricesVector);
408  computeCovariances<PointSource>(input_, tree_reciprocal_, *input_covariances_);
409  }
410 
411  base_transformation_ = Eigen::Matrix4f::Identity();
412  nr_iterations_ = 0;
413  converged_ = false;
414  double dist_threshold = corr_dist_threshold_ * corr_dist_threshold_;
415  pcl::Indices nn_indices(1);
416  std::vector<float> nn_dists(1);
417 
418  pcl::transformPointCloud(output, output, guess);
419 
420  while (!converged_) {
421  std::size_t cnt = 0;
422  std::vector<int> source_indices(indices_->size());
423  std::vector<int> target_indices(indices_->size());
424 
425  // guess corresponds to base_t and transformation_ to t
426  Eigen::Matrix4d transform_R = Eigen::Matrix4d::Zero();
427  for (std::size_t i = 0; i < 4; i++)
428  for (std::size_t j = 0; j < 4; j++)
429  for (std::size_t k = 0; k < 4; k++)
430  transform_R(i, j) += double(transformation_(i, k)) * double(guess(k, j));
431 
432  Eigen::Matrix3d R = transform_R.topLeftCorner<3, 3>();
433 
434  for (std::size_t i = 0; i < N; i++) {
435  PointSource query = output[i];
436  query.getVector4fMap() = transformation_ * query.getVector4fMap();
437 
438  if (!searchForNeighbors(query, nn_indices, nn_dists)) {
439  PCL_ERROR("[pcl::%s::computeTransformation] Unable to find a nearest neighbor "
440  "in the target dataset for point %d in the source!\n",
441  getClassName().c_str(),
442  (*indices_)[i]);
443  return;
444  }
445 
446  // Check if the distance to the nearest neighbor is smaller than the user imposed
447  // threshold
448  if (nn_dists[0] < dist_threshold) {
449  Eigen::Matrix3d& C1 = (*input_covariances_)[i];
450  Eigen::Matrix3d& C2 = (*target_covariances_)[nn_indices[0]];
451  Eigen::Matrix3d& M = mahalanobis_[i];
452  // M = R*C1
453  M = R * C1;
454  // temp = M*R' + C2 = R*C1*R' + C2
455  Eigen::Matrix3d temp = M * R.transpose();
456  temp += C2;
457  // M = temp^-1
458  M = temp.inverse();
459  source_indices[cnt] = static_cast<int>(i);
460  target_indices[cnt] = nn_indices[0];
461  cnt++;
462  }
463  }
464  // Resize to the actual number of valid correspondences
465  source_indices.resize(cnt);
466  target_indices.resize(cnt);
467  /* optimize transformation using the current assignment and Mahalanobis metrics*/
468  previous_transformation_ = transformation_;
469  // optimization right here
470  try {
471  rigid_transformation_estimation_(
472  output, source_indices, *target_, target_indices, transformation_);
473  /* compute the delta from this iteration */
474  delta = 0.;
475  for (int k = 0; k < 4; k++) {
476  for (int l = 0; l < 4; l++) {
477  double ratio = 1;
478  if (k < 3 && l < 3) // rotation part of the transform
479  ratio = 1. / rotation_epsilon_;
480  else
481  ratio = 1. / transformation_epsilon_;
482  double c_delta =
483  ratio * std::abs(previous_transformation_(k, l) - transformation_(k, l));
484  if (c_delta > delta)
485  delta = c_delta;
486  }
487  }
488  } catch (PCLException& e) {
489  PCL_DEBUG("[pcl::%s::computeTransformation] Optimization issue %s\n",
490  getClassName().c_str(),
491  e.what());
492  break;
493  }
494  nr_iterations_++;
495  // Check for convergence
496  if (nr_iterations_ >= max_iterations_ || delta < 1) {
497  converged_ = true;
498  PCL_DEBUG("[pcl::%s::computeTransformation] Convergence reached. Number of "
499  "iterations: %d out of %d. Transformation difference: %f\n",
500  getClassName().c_str(),
501  nr_iterations_,
502  max_iterations_,
503  (transformation_ - previous_transformation_).array().abs().sum());
504  previous_transformation_ = transformation_;
505  }
506  else
507  PCL_DEBUG("[pcl::%s::computeTransformation] Convergence failed\n",
508  getClassName().c_str());
509  }
510  final_transformation_ = previous_transformation_ * guess;
511 
512  // Transform the point cloud
513  pcl::transformPointCloud(*input_, output, final_transformation_);
514 }
515 
516 template <typename PointSource, typename PointTarget>
517 void
518 GeneralizedIterativeClosestPoint<PointSource, PointTarget>::applyState(
519  Eigen::Matrix4f& t, const Vector6d& x) const
520 {
521  // Z Y X euler angles convention
522  Eigen::Matrix3f R;
523  R = Eigen::AngleAxisf(static_cast<float>(x[5]), Eigen::Vector3f::UnitZ()) *
524  Eigen::AngleAxisf(static_cast<float>(x[4]), Eigen::Vector3f::UnitY()) *
525  Eigen::AngleAxisf(static_cast<float>(x[3]), Eigen::Vector3f::UnitX());
526  t.topLeftCorner<3, 3>().matrix() = R * t.topLeftCorner<3, 3>().matrix();
527  Eigen::Vector4f T(static_cast<float>(x[0]),
528  static_cast<float>(x[1]),
529  static_cast<float>(x[2]),
530  0.0f);
531  t.col(3) += T;
532 }
533 
534 } // namespace pcl
535 
536 #endif // PCL_REGISTRATION_IMPL_GICP_HPP_
pcl::GeneralizedIterativeClosestPoint::computeRDerivative
void computeRDerivative(const Vector6d &x, const Eigen::Matrix3d &R, Vector6d &g) const
Computes rotation matrix derivative.
Definition: gicp.hpp:131
pcl
Definition: convolution.h:46
BFGSSpace::NegativeGradientEpsilon
@ NegativeGradientEpsilon
Definition: bfgs.h:71
pcl::search::KdTree::nearestKSearch
int nearestKSearch(const PointT &point, int k, Indices &k_indices, std::vector< float > &k_sqr_distances) const override
Search for the k-nearest neighbors for the given query point.
Definition: kdtree.hpp:87
pcl::GeneralizedIterativeClosestPoint< PointXYZRGBA, PointXYZRGBA >::Vector6d
Eigen::Matrix< double, 6, 1 > Vector6d
Definition: gicp.h:103
pcl::NotEnoughPointsException
An exception that is thrown when the number of correspondents is not equal to the minimum required.
Definition: exceptions.h:63
pcl::GeneralizedIterativeClosestPoint::estimateRigidTransformationBFGS
void estimateRigidTransformationBFGS(const PointCloudSource &cloud_src, const std::vector< int > &indices_src, const PointCloudTarget &cloud_tgt, const std::vector< int > &indices_tgt, Eigen::Matrix4f &transformation_matrix)
Estimate a rigid rotation transformation between a source and a target point cloud using an iterative...
Definition: gicp.hpp:188
pcl::GeneralizedIterativeClosestPoint::applyState
void applyState(Eigen::Matrix4f &t, const Vector6d &x) const
compute transformation matrix from transformation matrix
Definition: gicp.hpp:518
pcl::PointCloud::begin
iterator begin() noexcept
Definition: point_cloud.h:423
BFGSSpace::NoProgress
@ NoProgress
Definition: bfgs.h:75
BFGS::minimizeInit
BFGSSpace::Status minimizeInit(FVectorType &x)
Definition: bfgs.h:363
pcl::GeneralizedIterativeClosestPoint::computeTransformation
void computeTransformation(PointCloudSource &output, const Eigen::Matrix4f &guess) override
Rigid transformation computation method with initial guess.
Definition: gicp.hpp:390
pcl::Registration< PointSource, PointTarget, float >::initComputeReciprocal
bool initComputeReciprocal()
Internal computation when reciprocal lookup is needed.
Definition: registration.hpp:105
pcl::PointCloud< PointXYZRGBA >
BFGS::minimizeOneStep
BFGSSpace::Status minimizeOneStep(FVectorType &x)
Definition: bfgs.h:395
pcl::PointXYZRGB
A point structure representing Euclidean xyz coordinates, and the RGB color.
Definition: point_types.hpp:630
BFGS::testGradient
BFGSSpace::Status testGradient()
Definition: bfgs.h:478
pcl::transformPointCloud
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
pcl::search::KdTree::Ptr
shared_ptr< KdTree< PointT, Tree > > Ptr
Definition: kdtree.h:75
BFGS
BFGS stands for Broyden–Fletcher–Goldfarb–Shanno (BFGS) method for solving unconstrained nonlinear op...
Definition: bfgs.h:121
BFGSSpace::Running
@ Running
Definition: bfgs.h:73
pcl::SolverDidntConvergeException
An exception that is thrown when the non linear solver didn't converge.
Definition: exceptions.h:49
pcl::PointCloud::end
iterator end() noexcept
Definition: point_cloud.h:424
pcl::GeneralizedIterativeClosestPoint::OptimizationFunctorWithIndices::operator()
double operator()(const Vector6d &x) override
Definition: gicp.hpp:261
pcl::GeneralizedIterativeClosestPoint::computeCovariances
void computeCovariances(typename pcl::PointCloud< PointT >::ConstPtr cloud, const typename pcl::search::KdTree< PointT >::Ptr tree, MatricesVector &cloud_covariances)
compute points covariances matrices according to the K nearest neighbors.
Definition: gicp.hpp:51
pcl::Indices
IndicesAllocator<> Indices
Type used for indices in PCL.
Definition: types.h:133
pcl::PointCloud::size
std::size_t size() const
Definition: point_cloud.h:437
pcl::PCLException
A base class for all pcl exceptions which inherits from std::runtime_error.
Definition: exceptions.h:63
BFGSSpace::Success
@ Success
Definition: bfgs.h:74
BFGSSpace::Status
Status
Definition: bfgs.h:70
pcl::PointCloud::ConstPtr
shared_ptr< const PointCloud< PointT > > ConstPtr
Definition: point_cloud.h:408
pcl::GeneralizedIterativeClosestPoint< PointXYZRGBA, PointXYZRGBA >::MatricesVector
std::vector< Eigen::Matrix3d, Eigen::aligned_allocator< Eigen::Matrix3d > > MatricesVector
Definition: gicp.h:92
pcl::GeneralizedIterativeClosestPoint::OptimizationFunctorWithIndices::df
void df(const Vector6d &x, Vector6d &df) override
Definition: gicp.hpp:289
pcl::GeneralizedIterativeClosestPoint::OptimizationFunctorWithIndices::checkGradient
BFGSSpace::Status checkGradient(const Vector6d &g) override
Definition: gicp.hpp:368
pcl::Vector4fMapConst
const Eigen::Map< const Eigen::Vector4f, Eigen::Aligned > Vector4fMapConst
Definition: point_types.hpp:185
pcl::GeneralizedIterativeClosestPoint::OptimizationFunctorWithIndices::fdf
void fdf(const Vector6d &x, double &f, Vector6d &df) override
Definition: gicp.hpp:328
BFGS::parameters
Parameters parameters
Definition: bfgs.h:171