Point Cloud Library (PCL)  1.14.1-dev
organized_multi_plane_segmentation.hpp
1 /*
2  * Software License Agreement (BSD License)
3  *
4  * Point Cloud Library (PCL) - www.pointclouds.org
5  * Copyright (c) 2010-2012, Willow Garage, 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
11  * are met:
12  *
13  * * Redistributions of source code must retain the above copyright
14  * notice, this list of conditions and the following disclaimer.
15  * * Redistributions in binary form must reproduce the above
16  * copyright notice, this list of conditions and the following
17  * disclaimer in the documentation and/or other materials provided
18  * with the distribution.
19  * * Neither the name of the copyright holder(s) nor the names of its
20  * contributors may be used to endorse or promote products derived
21  * from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
27  * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
29  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
31  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
33  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
34  * POSSIBILITY OF SUCH DAMAGE.
35  *
36  *
37  *
38  */
39 
40 #ifndef PCL_SEGMENTATION_IMPL_ORGANIZED_MULTI_PLANE_SEGMENTATION_H_
41 #define PCL_SEGMENTATION_IMPL_ORGANIZED_MULTI_PLANE_SEGMENTATION_H_
42 
43 #include <pcl/segmentation/organized_connected_component_segmentation.h>
44 #include <pcl/segmentation/organized_multi_plane_segmentation.h>
45 #include <pcl/common/centroid.h>
46 #include <pcl/common/eigen.h>
47 
48 /////////////////////////////////////////////////////////////////////////////////////////////////////////////////
49 template <typename PointT> pcl::PointCloud<PointT>
50 projectToPlaneFromViewpoint (pcl::PointCloud<PointT>& cloud, Eigen::Vector4f& normal, Eigen::Vector3f& centroid, Eigen::Vector3f& vp)
51 {
52  Eigen::Vector3f norm (normal[0], normal[1], normal[2]); //(region.coefficients_[0], region.coefficients_[1], region.coefficients_[2]);
53  pcl::PointCloud<PointT> projected_cloud;
54  projected_cloud.resize (cloud.size ());
55  for (std::size_t i = 0; i < cloud.size (); i++)
56  {
57  Eigen::Vector3f pt (cloud[i].x, cloud[i].y, cloud[i].z);
58  //Eigen::Vector3f intersection = (vp, pt, norm, centroid);
59  float u = norm.dot ((centroid - vp)) / norm.dot ((pt - vp));
60  Eigen::Vector3f intersection (vp + u * (pt - vp));
61  projected_cloud[i].x = intersection[0];
62  projected_cloud[i].y = intersection[1];
63  projected_cloud[i].z = intersection[2];
64  }
65 
66  return (projected_cloud);
67 }
68 
69 //////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
70 template<typename PointT, typename PointNT, typename PointLT> void
71 pcl::OrganizedMultiPlaneSegmentation<PointT, PointNT, PointLT>::segment (std::vector<ModelCoefficients>& model_coefficients,
72  std::vector<PointIndices>& inlier_indices)
73 {
75  std::vector<pcl::PointIndices> label_indices;
76  std::vector<Eigen::Vector4f, Eigen::aligned_allocator<Eigen::Vector4f> > centroids;
77  std::vector <Eigen::Matrix3f, Eigen::aligned_allocator<Eigen::Matrix3f> > covariances;
78  segment (model_coefficients, inlier_indices, centroids, covariances, labels, label_indices);
79 }
80 
81 //////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
82 template<typename PointT, typename PointNT, typename PointLT> void
83 pcl::OrganizedMultiPlaneSegmentation<PointT, PointNT, PointLT>::segment (std::vector<ModelCoefficients>& model_coefficients,
84  std::vector<PointIndices>& inlier_indices,
85  std::vector<Eigen::Vector4f, Eigen::aligned_allocator<Eigen::Vector4f> >& centroids,
86  std::vector <Eigen::Matrix3f, Eigen::aligned_allocator<Eigen::Matrix3f> >& covariances,
88  std::vector<pcl::PointIndices>& label_indices)
89 {
90  if (!initCompute ())
91  return;
92 
93  // Check that the normals are present
94  if (!normals_)
95  {
96  PCL_ERROR( "[pcl::%s::segment] Must specify normals.\n", getClassName().c_str());
97  return;
98  }
99 
100  // Check that we got the same number of points and normals
101  if (normals_->size () != input_->size ())
102  {
103  PCL_ERROR("[pcl::%s::segment] Number of points in input cloud (%zu) and normal "
104  "cloud (%zu) do not match!\n",
105  getClassName().c_str(),
106  static_cast<std::size_t>(input_->size()),
107  static_cast<std::size_t>(normals_->size()));
108  return;
109  }
110 
111  // Check that the cloud is organized
112  if (!input_->isOrganized ())
113  {
114  PCL_ERROR ("[pcl::%s::segment] Organized point cloud is required for this plane extraction method!\n",
115  getClassName ().c_str ());
116  return;
117  }
118 
119  // Calculate range part of planes' hessian normal form
120  std::vector<float> plane_d (input_->size ());
121 
122  for (std::size_t i = 0; i < input_->size (); ++i)
123  plane_d[i] = (*input_)[i].getVector3fMap ().dot ((*normals_)[i].getNormalVector3fMap ());
124 
125  // Make a comparator
126  //PlaneCoefficientComparator<PointT,PointNT> plane_comparator (plane_d);
127  compare_->setPlaneCoeffD (plane_d);
128  compare_->setInputCloud (input_);
129  compare_->setInputNormals (normals_);
130  compare_->setAngularThreshold (static_cast<float> (angular_threshold_));
131  compare_->setDistanceThreshold (static_cast<float> (distance_threshold_), true);
132 
133  // Set up the output
134  OrganizedConnectedComponentSegmentation<PointT,PointLT> connected_component (compare_);
135  connected_component.setInputCloud (input_);
136  connected_component.segment (labels, label_indices);
137 
138  Eigen::Vector4f clust_centroid = Eigen::Vector4f::Zero ();
139  Eigen::Vector4f vp = Eigen::Vector4f::Zero ();
140  Eigen::Matrix3f clust_cov;
142  model.values.resize (4);
143 
144  // Fit Planes to each cluster
145  for (const auto &label_index : label_indices)
146  {
147  if (static_cast<unsigned> (label_index.indices.size ()) > min_inliers_)
148  {
149  pcl::computeMeanAndCovarianceMatrix (*input_, label_index.indices, clust_cov, clust_centroid);
150  Eigen::Vector4f plane_params;
151 
152  EIGEN_ALIGN16 Eigen::Vector3f::Scalar eigen_value;
153  EIGEN_ALIGN16 Eigen::Vector3f eigen_vector;
154  pcl::eigen33 (clust_cov, eigen_value, eigen_vector);
155  plane_params[0] = eigen_vector[0];
156  plane_params[1] = eigen_vector[1];
157  plane_params[2] = eigen_vector[2];
158  plane_params[3] = 0;
159  plane_params[3] = -1 * plane_params.dot (clust_centroid);
160 
161  vp -= clust_centroid;
162  float cos_theta = vp.dot (plane_params);
163  if (cos_theta < 0)
164  {
165  plane_params *= -1;
166  plane_params[3] = 0;
167  plane_params[3] = -1 * plane_params.dot (clust_centroid);
168  }
169 
170  // Compute the curvature surface change
171  float curvature;
172  float eig_sum = clust_cov.coeff (0) + clust_cov.coeff (4) + clust_cov.coeff (8);
173  if (eig_sum != 0)
174  curvature = std::abs (eigen_value / eig_sum);
175  else
176  curvature = 0;
177 
178  if (curvature < maximum_curvature_)
179  {
180  model.values[0] = plane_params[0];
181  model.values[1] = plane_params[1];
182  model.values[2] = plane_params[2];
183  model.values[3] = plane_params[3];
184  model_coefficients.push_back (model);
185  inlier_indices.push_back (label_index);
186  centroids.push_back (clust_centroid);
187  covariances.push_back (clust_cov);
188  }
189  }
190  }
191  deinitCompute ();
192 }
193 
194 //////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
195 template<typename PointT, typename PointNT, typename PointLT> void
197 {
198  std::vector<ModelCoefficients> model_coefficients;
199  std::vector<PointIndices> inlier_indices;
200  PointCloudLPtr labels (new PointCloudL);
201  std::vector<pcl::PointIndices> label_indices;
202  std::vector<pcl::PointIndices> boundary_indices;
203  pcl::PointCloud<PointT> boundary_cloud;
204  std::vector<Eigen::Vector4f, Eigen::aligned_allocator<Eigen::Vector4f> > centroids;
205  std::vector <Eigen::Matrix3f, Eigen::aligned_allocator<Eigen::Matrix3f> > covariances;
206  segment (model_coefficients, inlier_indices, centroids, covariances, *labels, label_indices);
207  regions.resize (model_coefficients.size ());
208  boundary_indices.resize (model_coefficients.size ());
209 
210  for (std::size_t i = 0; i < model_coefficients.size (); i++)
211  {
212  boundary_cloud.resize (0);
213  pcl::OrganizedConnectedComponentSegmentation<PointT,PointLT>::findLabeledRegionBoundary (inlier_indices[i].indices[0], labels, boundary_indices[i]);
214  boundary_cloud.resize (boundary_indices[i].indices.size ());
215  for (std::size_t j = 0; j < boundary_indices[i].indices.size (); j++)
216  boundary_cloud[j] = (*input_)[boundary_indices[i].indices[j]];
217 
218  Eigen::Vector3f centroid = Eigen::Vector3f (centroids[i][0],centroids[i][1],centroids[i][2]);
219  Eigen::Vector4f model = Eigen::Vector4f (model_coefficients[i].values[0],
220  model_coefficients[i].values[1],
221  model_coefficients[i].values[2],
222  model_coefficients[i].values[3]);
223  regions[i] = PlanarRegion<PointT> (centroid,
224  covariances[i],
225  static_cast<unsigned int> (inlier_indices[i].indices.size ()),
226  boundary_cloud.points,
227  model);
228  }
229 }
230 
231 //////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
232 template<typename PointT, typename PointNT, typename PointLT> void
234 {
235  std::vector<ModelCoefficients> model_coefficients;
236  std::vector<PointIndices> inlier_indices;
237  PointCloudLPtr labels (new PointCloudL);
238  std::vector<pcl::PointIndices> label_indices;
239  std::vector<pcl::PointIndices> boundary_indices;
240  pcl::PointCloud<PointT> boundary_cloud;
241  std::vector<Eigen::Vector4f, Eigen::aligned_allocator<Eigen::Vector4f> > centroids;
242  std::vector <Eigen::Matrix3f, Eigen::aligned_allocator<Eigen::Matrix3f> > covariances;
243  segment (model_coefficients, inlier_indices, centroids, covariances, *labels, label_indices);
244  refine (model_coefficients, inlier_indices, labels, label_indices);
245  regions.resize (model_coefficients.size ());
246  boundary_indices.resize (model_coefficients.size ());
247 
248  for (std::size_t i = 0; i < model_coefficients.size (); i++)
249  {
250  boundary_cloud.resize (0);
251  int max_inlier_idx = static_cast<int> (inlier_indices[i].indices.size ()) - 1;
252  pcl::OrganizedConnectedComponentSegmentation<PointT,PointLT>::findLabeledRegionBoundary (inlier_indices[i].indices[max_inlier_idx], labels, boundary_indices[i]);
253  boundary_cloud.resize (boundary_indices[i].indices.size ());
254  for (std::size_t j = 0; j < boundary_indices[i].indices.size (); j++)
255  boundary_cloud[j] = (*input_)[boundary_indices[i].indices[j]];
256 
257  Eigen::Vector3f centroid = Eigen::Vector3f (centroids[i][0],centroids[i][1],centroids[i][2]);
258  Eigen::Vector4f model = Eigen::Vector4f (model_coefficients[i].values[0],
259  model_coefficients[i].values[1],
260  model_coefficients[i].values[2],
261  model_coefficients[i].values[3]);
262 
263  Eigen::Vector3f vp (0.0, 0.0, 0.0);
264  if (project_points_)
265  boundary_cloud = projectToPlaneFromViewpoint (boundary_cloud, model, centroid, vp);
266 
267  regions[i] = PlanarRegion<PointT> (centroid,
268  covariances[i],
269  static_cast<unsigned int> (inlier_indices[i].indices.size ()),
270  boundary_cloud.points,
271  model);
272  }
273 }
274 
275 //////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
276 template<typename PointT, typename PointNT, typename PointLT> void
278  std::vector<ModelCoefficients>& model_coefficients,
279  std::vector<PointIndices>& inlier_indices,
280  PointCloudLPtr& labels,
281  std::vector<pcl::PointIndices>& label_indices,
282  std::vector<pcl::PointIndices>& boundary_indices)
283 {
284  pcl::PointCloud<PointT> boundary_cloud;
285  std::vector<Eigen::Vector4f, Eigen::aligned_allocator<Eigen::Vector4f> > centroids;
286  std::vector <Eigen::Matrix3f, Eigen::aligned_allocator<Eigen::Matrix3f> > covariances;
287  segment (model_coefficients, inlier_indices, centroids, covariances, *labels, label_indices);
288  refine (model_coefficients, inlier_indices, labels, label_indices);
289  regions.resize (model_coefficients.size ());
290  boundary_indices.resize (model_coefficients.size ());
291 
292  for (std::size_t i = 0; i < model_coefficients.size (); i++)
293  {
294  boundary_cloud.resize (0);
295  int max_inlier_idx = static_cast<int> (inlier_indices[i].indices.size ()) - 1;
296  pcl::OrganizedConnectedComponentSegmentation<PointT,PointLT>::findLabeledRegionBoundary (inlier_indices[i].indices[max_inlier_idx], labels, boundary_indices[i]);
297  boundary_cloud.resize (boundary_indices[i].indices.size ());
298  for (std::size_t j = 0; j < boundary_indices[i].indices.size (); j++)
299  boundary_cloud[j] = (*input_)[boundary_indices[i].indices[j]];
300 
301  Eigen::Vector3f centroid = Eigen::Vector3f (centroids[i][0],centroids[i][1],centroids[i][2]);
302  Eigen::Vector4f model = Eigen::Vector4f (model_coefficients[i].values[0],
303  model_coefficients[i].values[1],
304  model_coefficients[i].values[2],
305  model_coefficients[i].values[3]);
306 
307  Eigen::Vector3f vp (0.0, 0.0, 0.0);
308  if (project_points_ && !boundary_cloud.empty ())
309  boundary_cloud = projectToPlaneFromViewpoint (boundary_cloud, model, centroid, vp);
310 
311  regions[i] = PlanarRegion<PointT> (centroid,
312  covariances[i],
313  static_cast<unsigned int> (inlier_indices[i].indices.size ()),
314  boundary_cloud.points,
315  model);
316  }
317 }
318 
319 //////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
320 template<typename PointT, typename PointNT, typename PointLT> void
321 pcl::OrganizedMultiPlaneSegmentation<PointT, PointNT, PointLT>::refine (std::vector<ModelCoefficients>& model_coefficients,
322  std::vector<PointIndices>& inlier_indices,
323  PointCloudLPtr& labels,
324  std::vector<pcl::PointIndices>& label_indices)
325 {
326  //List of labels to grow, and index of model corresponding to each label
327  std::vector<bool> grow_labels;
328  std::vector<int> label_to_model;
329  grow_labels.resize (label_indices.size (), false);
330  label_to_model.resize (label_indices.size (), 0);
331 
332  for (std::size_t i = 0; i < model_coefficients.size (); i++)
333  {
334  int model_label = (*labels)[inlier_indices[i].indices[0]].label;
335  label_to_model[model_label] = static_cast<int> (i);
336  grow_labels[model_label] = true;
337  }
338 
339  //refinement_compare_->setDistanceThreshold (0.015f, true);
340  refinement_compare_->setInputCloud (input_);
341  refinement_compare_->setLabels (labels);
342  refinement_compare_->setModelCoefficients (model_coefficients);
343  refinement_compare_->setRefineLabels (grow_labels);
344  refinement_compare_->setLabelToModel (label_to_model);
345 
346  //Do a first pass over the image, top to bottom, left to right
347  unsigned int current_row = 0;
348  unsigned int next_row = labels->width;
349  for (std::size_t rowIdx = 0; rowIdx < labels->height - 1; ++rowIdx, current_row = next_row, next_row += labels->width)
350  {
351 
352  for (unsigned colIdx = 0; colIdx < labels->width - 1; ++colIdx)
353  {
354  int current_label = (*labels)[current_row+colIdx].label;
355  int right_label = (*labels)[current_row+colIdx+1].label;
356  if (current_label < 0 || right_label < 0)
357  continue;
358 
359  //Check right
360  //bool test1 = false;
361  if (refinement_compare_->compare (current_row+colIdx, current_row+colIdx+1))
362  {
363  //test1 = true;
364  (*labels)[current_row+colIdx+1].label = current_label;
365  label_indices[current_label].indices.push_back (current_row+colIdx+1);
366  inlier_indices[label_to_model[current_label]].indices.push_back (current_row+colIdx+1);
367  }
368 
369  int lower_label = (*labels)[next_row+colIdx].label;
370  if (lower_label < 0)
371  continue;
372 
373  //Check down
374  if (refinement_compare_->compare (current_row+colIdx, next_row+colIdx))
375  {
376  (*labels)[next_row+colIdx].label = current_label;
377  label_indices[current_label].indices.push_back (next_row+colIdx);
378  inlier_indices[label_to_model[current_label]].indices.push_back (next_row+colIdx);
379  }
380 
381  }//col
382  }//row
383 
384  //Do a second pass over the image
385  current_row = labels->width * (labels->height - 1);
386  unsigned int prev_row = current_row - labels->width;
387  for (std::size_t rowIdx = 0; rowIdx < labels->height - 1; ++rowIdx, current_row = prev_row, prev_row -= labels->width)
388  {
389  for (int colIdx = labels->width - 1; colIdx >= 0; --colIdx)
390  {
391  int current_label = (*labels)[current_row+colIdx].label;
392  int left_label = (*labels)[current_row+colIdx-1].label;
393  if (current_label < 0 || left_label < 0)
394  continue;
395 
396  //Check left
397  if (refinement_compare_->compare (current_row+colIdx, current_row+colIdx-1))
398  {
399  (*labels)[current_row+colIdx-1].label = current_label;
400  label_indices[current_label].indices.push_back (current_row+colIdx-1);
401  inlier_indices[label_to_model[current_label]].indices.push_back (current_row+colIdx-1);
402  }
403 
404  int upper_label = (*labels)[prev_row+colIdx].label;
405  if (upper_label < 0)
406  continue;
407  //Check up
408  if (refinement_compare_->compare (current_row+colIdx, prev_row+colIdx))
409  {
410  (*labels)[prev_row+colIdx].label = current_label;
411  label_indices[current_label].indices.push_back (prev_row+colIdx);
412  inlier_indices[label_to_model[current_label]].indices.push_back (prev_row+colIdx);
413  }
414  }//col
415  }//row
416 }
417 
418 #define PCL_INSTANTIATE_OrganizedMultiPlaneSegmentation(T,NT,LT) template class PCL_EXPORTS pcl::OrganizedMultiPlaneSegmentation<T,NT,LT>;
419 
420 #endif // PCL_SEGMENTATION_IMPL_MULTI_PLANE_SEGMENTATION_H_
Define methods for centroid estimation and covariance matrix calculus.
OrganizedConnectedComponentSegmentation allows connected components to be found within organized poin...
static void findLabeledRegionBoundary(int start_idx, PointCloudLPtr labels, pcl::PointIndices &boundary_indices)
Find the boundary points / contour of a connected component.
void segment(pcl::PointCloud< PointLT > &labels, std::vector< pcl::PointIndices > &label_indices) const
Perform the connected component segmentation.
void segment(std::vector< ModelCoefficients > &model_coefficients, std::vector< PointIndices > &inlier_indices, std::vector< Eigen::Vector4f, Eigen::aligned_allocator< Eigen::Vector4f > > &centroids, std::vector< Eigen::Matrix3f, Eigen::aligned_allocator< Eigen::Matrix3f > > &covariances, pcl::PointCloud< PointLT > &labels, std::vector< pcl::PointIndices > &label_indices)
Segmentation of all planes in a point cloud given by setInputCloud(), setIndices()
void segmentAndRefine(std::vector< PlanarRegion< PointT >, Eigen::aligned_allocator< PlanarRegion< PointT > > > &regions)
Perform a segmentation, as well as an additional refinement step.
void refine(std::vector< ModelCoefficients > &model_coefficients, std::vector< PointIndices > &inlier_indices, PointCloudLPtr &labels, std::vector< pcl::PointIndices > &label_indices)
Perform a refinement of an initial segmentation, by comparing points to adjacent regions detected by ...
virtual void setInputCloud(const PointCloudConstPtr &cloud)
Provide a pointer to the input dataset.
Definition: pcl_base.hpp:65
PlanarRegion represents a set of points that lie in a plane.
Definition: planar_region.h:52
PointCloud represents the base class in PCL for storing collections of 3D points.
Definition: point_cloud.h:173
bool empty() const
Definition: point_cloud.h:446
void resize(std::size_t count)
Resizes the container to contain count elements.
Definition: point_cloud.h:462
std::size_t size() const
Definition: point_cloud.h:443
std::vector< PointT, Eigen::aligned_allocator< PointT > > points
The point data.
Definition: point_cloud.h:395
unsigned int computeMeanAndCovarianceMatrix(const pcl::PointCloud< PointT > &cloud, Eigen::Matrix< Scalar, 3, 3 > &covariance_matrix, Eigen::Matrix< Scalar, 4, 1 > &centroid)
Compute the normalized 3x3 covariance matrix and the centroid of a given set of points in a single lo...
Definition: centroid.hpp:509
void eigen33(const Matrix &mat, typename Matrix::Scalar &eigenvalue, Vector &eigenvector)
determines the eigenvector and eigenvalue of the smallest eigenvalue of the symmetric positive semi d...
Definition: eigen.hpp:295
__device__ __host__ __forceinline__ float norm(const float3 &v1, const float3 &v2)
std::vector< float > values