fpfh.hpp
Go to the documentation of this file.
00001 /*
00002  * Software License Agreement (BSD License)
00003  *
00004  *  Point Cloud Library (PCL) - www.pointclouds.org
00005  *  Copyright (c) 2010-2011, Willow Garage, Inc.
00006  *
00007  *  All rights reserved.
00008  *
00009  *  Redistribution and use in source and binary forms, with or without
00010  *  modification, are permitted provided that the following conditions
00011  *  are met:
00012  *
00013  *   * Redistributions of source code must retain the above copyright
00014  *     notice, this list of conditions and the following disclaimer.
00015  *   * Redistributions in binary form must reproduce the above
00016  *     copyright notice, this list of conditions and the following
00017  *     disclaimer in the documentation and/or other materials provided
00018  *     with the distribution.
00019  *   * Neither the name of Willow Garage, Inc. nor the names of its
00020  *     contributors may be used to endorse or promote products derived
00021  *     from this software without specific prior written permission.
00022  *
00023  *  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
00024  *  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
00025  *  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
00026  *  FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
00027  *  COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
00028  *  INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
00029  *  BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
00030  *  LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
00031  *  CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
00032  *  LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
00033  *  ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
00034  *  POSSIBILITY OF SUCH DAMAGE.
00035  *
00036  * $Id: fpfh.hpp 4927 2012-03-07 03:35:53Z rusu $
00037  *
00038  */
00039 
00040 #ifndef PCL_FEATURES_IMPL_FPFH_H_
00041 #define PCL_FEATURES_IMPL_FPFH_H_
00042 
00043 #include <pcl/features/fpfh.h>
00044 #include <pcl/features/pfh.h>
00045 
00047 template <typename PointInT, typename PointNT, typename PointOutT> bool
00048 pcl::FPFHEstimation<PointInT, PointNT, PointOutT>::computePairFeatures (
00049     const pcl::PointCloud<PointInT> &cloud, const pcl::PointCloud<PointNT> &normals,
00050     int p_idx, int q_idx, float &f1, float &f2, float &f3, float &f4)
00051 {
00052   pcl::computePairFeatures (cloud.points[p_idx].getVector4fMap (), normals.points[p_idx].getNormalVector4fMap (),
00053       cloud.points[q_idx].getVector4fMap (), normals.points[q_idx].getNormalVector4fMap (),
00054       f1, f2, f3, f4);
00055   return (true);
00056 }
00057 
00059 template <typename PointInT, typename PointNT, typename PointOutT> void 
00060 pcl::FPFHEstimation<PointInT, PointNT, PointOutT>::computePointSPFHSignature (
00061     const pcl::PointCloud<PointInT> &cloud, const pcl::PointCloud<PointNT> &normals,
00062     int p_idx, int row, const std::vector<int> &indices,
00063     Eigen::MatrixXf &hist_f1, Eigen::MatrixXf &hist_f2, Eigen::MatrixXf &hist_f3)
00064 {
00065   Eigen::Vector4f pfh_tuple;
00066   // Get the number of bins from the histograms size
00067   int nr_bins_f1 = static_cast<int> (hist_f1.cols ());
00068   int nr_bins_f2 = static_cast<int> (hist_f2.cols ());
00069   int nr_bins_f3 = static_cast<int> (hist_f3.cols ());
00070 
00071   // Factorization constant
00072   float hist_incr = 100.0f / static_cast<float>(indices.size () - 1);
00073 
00074   // Iterate over all the points in the neighborhood
00075   for (size_t idx = 0; idx < indices.size (); ++idx)
00076   {
00077     // Avoid unnecessary returns
00078     if (p_idx == indices[idx])
00079         continue;
00080 
00081     // Compute the pair P to NNi
00082     if (!computePairFeatures (cloud, normals, p_idx, indices[idx], pfh_tuple[0], pfh_tuple[1], pfh_tuple[2], pfh_tuple[3]))
00083         continue;
00084 
00085     // Normalize the f1, f2, f3 features and push them in the histogram
00086     int h_index = static_cast<int> (floor (nr_bins_f1 * ((pfh_tuple[0] + M_PI) * d_pi_)));
00087     if (h_index < 0)           h_index = 0;
00088     if (h_index >= nr_bins_f1) h_index = nr_bins_f1 - 1;
00089     hist_f1 (row, h_index) += hist_incr;
00090 
00091     h_index = static_cast<int> (floor (nr_bins_f2 * ((pfh_tuple[1] + 1.0) * 0.5)));
00092     if (h_index < 0)           h_index = 0;
00093     if (h_index >= nr_bins_f2) h_index = nr_bins_f2 - 1;
00094     hist_f2 (row, h_index) += hist_incr;
00095 
00096     h_index = static_cast<int> (floor (nr_bins_f3 * ((pfh_tuple[2] + 1.0) * 0.5)));
00097     if (h_index < 0)           h_index = 0;
00098     if (h_index >= nr_bins_f3) h_index = nr_bins_f3 - 1;
00099     hist_f3 (row, h_index) += hist_incr;
00100   }
00101 }
00102 
00104 template <typename PointInT, typename PointNT, typename PointOutT> void
00105 pcl::FPFHEstimation<PointInT, PointNT, PointOutT>::weightPointSPFHSignature (
00106     const Eigen::MatrixXf &hist_f1, const Eigen::MatrixXf &hist_f2, const Eigen::MatrixXf &hist_f3,
00107     const std::vector<int> &indices, const std::vector<float> &dists, Eigen::VectorXf &fpfh_histogram)
00108 {
00109   assert (indices.size () == dists.size ());
00110   double sum_f1 = 0.0, sum_f2 = 0.0, sum_f3 = 0.0;
00111   float weight = 0.0, val_f1, val_f2, val_f3;
00112 
00113   // Get the number of bins from the histograms size
00114   int nr_bins_f1 = static_cast<int> (hist_f1.cols ());
00115   int nr_bins_f2 = static_cast<int> (hist_f2.cols ());
00116   int nr_bins_f3 = static_cast<int> (hist_f3.cols ());
00117   int nr_bins_f12 = nr_bins_f1 + nr_bins_f2;
00118 
00119   // Clear the histogram
00120   fpfh_histogram.setZero (nr_bins_f1 + nr_bins_f2 + nr_bins_f3);
00121 
00122   // Use the entire patch
00123   for (size_t idx = 0, data_size = indices.size (); idx < data_size; ++idx)
00124   {
00125     // Minus the query point itself
00126     if (dists[idx] == 0)
00127       continue;
00128 
00129     // Standard weighting function used
00130     weight = 1.0f / dists[idx];
00131 
00132     // Weight the SPFH of the query point with the SPFH of its neighbors
00133     for (int f1_i = 0; f1_i < nr_bins_f1; ++f1_i)
00134     {
00135       val_f1 = hist_f1 (indices[idx], f1_i) * weight;
00136       sum_f1 += val_f1;
00137       fpfh_histogram[f1_i] += val_f1;
00138     }
00139 
00140     for (int f2_i = 0; f2_i < nr_bins_f2; ++f2_i)
00141     {
00142       val_f2 = hist_f2 (indices[idx], f2_i) * weight;
00143       sum_f2 += val_f2;
00144       fpfh_histogram[f2_i + nr_bins_f1] += val_f2;
00145     }
00146 
00147     for (int f3_i = 0; f3_i < nr_bins_f3; ++f3_i)
00148     {
00149       val_f3 = hist_f3 (indices[idx], f3_i) * weight;
00150       sum_f3 += val_f3;
00151       fpfh_histogram[f3_i + nr_bins_f12] += val_f3;
00152     }
00153   }
00154 
00155   if (sum_f1 != 0)
00156     sum_f1 = 100.0 / sum_f1;           // histogram values sum up to 100
00157   if (sum_f2 != 0)
00158     sum_f2 = 100.0 / sum_f2;           // histogram values sum up to 100
00159   if (sum_f3 != 0)
00160     sum_f3 = 100.0 / sum_f3;           // histogram values sum up to 100
00161 
00162   // Adjust final FPFH values
00163   for (int f1_i = 0; f1_i < nr_bins_f1; ++f1_i)
00164     fpfh_histogram[f1_i] *= static_cast<float> (sum_f1);
00165   for (int f2_i = 0; f2_i < nr_bins_f2; ++f2_i)
00166     fpfh_histogram[f2_i + nr_bins_f1] *= static_cast<float> (sum_f2);
00167   for (int f3_i = 0; f3_i < nr_bins_f3; ++f3_i)
00168     fpfh_histogram[f3_i + nr_bins_f12] *= static_cast<float> (sum_f3);
00169 }
00170 
00172 template <typename PointInT, typename PointNT, typename PointOutT> void
00173 pcl::FPFHEstimation<PointInT, PointNT, PointOutT>::computeSPFHSignatures (std::vector<int> &spfh_hist_lookup,
00174     Eigen::MatrixXf &hist_f1, Eigen::MatrixXf &hist_f2, Eigen::MatrixXf &hist_f3)
00175 {
00176   // Allocate enough space to hold the NN search results
00177   // \note This resize is irrelevant for a radiusSearch ().
00178   std::vector<int> nn_indices (k_);
00179   std::vector<float> nn_dists (k_);
00180 
00181   std::set<int> spfh_indices;
00182   spfh_hist_lookup.resize (surface_->points.size ());
00183 
00184   // Build a list of (unique) indices for which we will need to compute SPFH signatures
00185   // (We need an SPFH signature for every point that is a neighbor of any point in input_[indices_])
00186   if (surface_ != input_ ||
00187       indices_->size () != surface_->points.size ())
00188   { 
00189     for (size_t idx = 0; idx < indices_->size (); ++idx)
00190     {
00191       int p_idx = (*indices_)[idx];
00192       if (this->searchForNeighbors (p_idx, search_parameter_, nn_indices, nn_dists) == 0)
00193         continue;
00194 
00195       spfh_indices.insert (nn_indices.begin (), nn_indices.end ());
00196     }
00197   }
00198   else
00199   {
00200     // Special case: When a feature must be computed at every point, there is no need for a neighborhood search
00201     for (size_t idx = 0; idx < indices_->size (); ++idx)
00202       spfh_indices.insert (static_cast<int> (idx));
00203   }
00204 
00205   // Initialize the arrays that will store the SPFH signatures
00206   size_t data_size = spfh_indices.size ();
00207   hist_f1.setZero (data_size, nr_bins_f1_);
00208   hist_f2.setZero (data_size, nr_bins_f2_);
00209   hist_f3.setZero (data_size, nr_bins_f3_);
00210 
00211   // Compute SPFH signatures for every point that needs them
00212   std::set<int>::iterator spfh_indices_itr = spfh_indices.begin ();
00213   for (int i = 0; i < static_cast<int> (spfh_indices.size ()); ++i)
00214   {
00215     // Get the next point index
00216     int p_idx = *spfh_indices_itr;
00217     ++spfh_indices_itr;
00218 
00219     // Find the neighborhood around p_idx
00220     if (this->searchForNeighbors (*surface_, p_idx, search_parameter_, nn_indices, nn_dists) == 0)
00221       continue;
00222 
00223     // Estimate the SPFH signature around p_idx
00224     computePointSPFHSignature (*surface_, *normals_, p_idx, i, nn_indices, hist_f1, hist_f2, hist_f3);
00225 
00226     // Populate a lookup table for converting a point index to its corresponding row in the spfh_hist_* matrices
00227     spfh_hist_lookup[p_idx] = i;
00228   }
00229 }
00230 
00232 template <typename PointInT, typename PointNT, typename PointOutT> void
00233 pcl::FPFHEstimation<PointInT, PointNT, PointOutT>::computeFeature (PointCloudOut &output)
00234 {
00235   // Allocate enough space to hold the NN search results
00236   // \note This resize is irrelevant for a radiusSearch ().
00237   std::vector<int> nn_indices (k_);
00238   std::vector<float> nn_dists (k_);
00239 
00240   std::vector<int> spfh_hist_lookup;
00241   computeSPFHSignatures (spfh_hist_lookup, hist_f1_, hist_f2_, hist_f3_);
00242 
00243   output.is_dense = true;
00244   // Save a few cycles by not checking every point for NaN/Inf values if the cloud is set to dense
00245   if (input_->is_dense)
00246   {
00247     // Iterate over the entire index vector
00248     for (size_t idx = 0; idx < indices_->size (); ++idx)
00249     {
00250       if (this->searchForNeighbors ((*indices_)[idx], search_parameter_, nn_indices, nn_dists) == 0)
00251       {
00252         for (int d = 0; d < fpfh_histogram_.size (); ++d)
00253           output.points[idx].histogram[d] = std::numeric_limits<float>::quiet_NaN ();
00254     
00255         output.is_dense = false;
00256         continue;
00257       }
00258 
00259       // ... and remap the nn_indices values so that they represent row indices in the spfh_hist_* matrices 
00260       // instead of indices into surface_->points
00261       for (size_t i = 0; i < nn_indices.size (); ++i)
00262         nn_indices[i] = spfh_hist_lookup[nn_indices[i]];
00263 
00264       // Compute the FPFH signature (i.e. compute a weighted combination of local SPFH signatures) ...
00265       weightPointSPFHSignature (hist_f1_, hist_f2_, hist_f3_, nn_indices, nn_dists, fpfh_histogram_);
00266 
00267       // ...and copy it into the output cloud
00268       for (int d = 0; d < fpfh_histogram_.size (); ++d)
00269         output.points[idx].histogram[d] = fpfh_histogram_[d];
00270     }
00271   }
00272   else
00273   {
00274     // Iterate over the entire index vector
00275     for (size_t idx = 0; idx < indices_->size (); ++idx)
00276     {
00277       if (!isFinite ((*input_)[(*indices_)[idx]]) ||
00278           this->searchForNeighbors ((*indices_)[idx], search_parameter_, nn_indices, nn_dists) == 0)
00279       {
00280         for (int d = 0; d < fpfh_histogram_.size (); ++d)
00281           output.points[idx].histogram[d] = std::numeric_limits<float>::quiet_NaN ();
00282     
00283         output.is_dense = false;
00284         continue;
00285       }
00286 
00287       // ... and remap the nn_indices values so that they represent row indices in the spfh_hist_* matrices 
00288       // instead of indices into surface_->points
00289       for (size_t i = 0; i < nn_indices.size (); ++i)
00290         nn_indices[i] = spfh_hist_lookup[nn_indices[i]];
00291 
00292       // Compute the FPFH signature (i.e. compute a weighted combination of local SPFH signatures) ...
00293       weightPointSPFHSignature (hist_f1_, hist_f2_, hist_f3_, nn_indices, nn_dists, fpfh_histogram_);
00294 
00295       // ...and copy it into the output cloud
00296       for (int d = 0; d < fpfh_histogram_.size (); ++d)
00297         output.points[idx].histogram[d] = fpfh_histogram_[d];
00298     }
00299   }
00300 }
00301 
00303 template <typename PointInT, typename PointNT> void
00304 pcl::FPFHEstimation<PointInT, PointNT, Eigen::MatrixXf>::computeFeatureEigen (pcl::PointCloud<Eigen::MatrixXf> &output)
00305 {
00306   // Set up the output channels
00307   output.channels["fpfh"].name     = "fpfh";
00308   output.channels["fpfh"].offset   = 0;
00309   output.channels["fpfh"].size     = 4;
00310   output.channels["fpfh"].count    = 33;
00311   output.channels["fpfh"].datatype = sensor_msgs::PointField::FLOAT32;
00312 
00313   // Allocate enough space to hold the NN search results
00314   // \note This resize is irrelevant for a radiusSearch ().
00315   std::vector<int> nn_indices (k_);
00316   std::vector<float> nn_dists (k_);
00317 
00318   std::vector<int> spfh_hist_lookup;
00319   this->computeSPFHSignatures (spfh_hist_lookup, hist_f1_, hist_f2_, hist_f3_);
00320 
00321   // Intialize the array that will store the FPFH signature
00322   output.points.resize (indices_->size (), nr_bins_f1_ + nr_bins_f2_ + nr_bins_f3_);
00323   output.is_dense = true;
00324   // Save a few cycles by not checking every point for NaN/Inf values if the cloud is set to dense
00325   if (input_->is_dense)
00326   {
00327     // Iterate over the entire index vector
00328     for (size_t idx = 0; idx < indices_->size (); ++idx)
00329     {
00330       if (!isFinite ((*input_)[(*indices_)[idx]]) ||
00331           this->searchForNeighbors ((*indices_)[idx], search_parameter_, nn_indices, nn_dists) == 0)
00332       {
00333         output.points.row (idx).setConstant (std::numeric_limits<float>::quiet_NaN ());
00334         output.is_dense = false;
00335         continue;
00336       }
00337 
00338       // ... and remap the nn_indices values so that they represent row indices in the spfh_hist_* matrices 
00339       // instead of indices into surface_->points
00340       for (size_t i = 0; i < nn_indices.size (); ++i)
00341         nn_indices[i] = spfh_hist_lookup[nn_indices[i]];
00342 
00343       // Compute the FPFH signature (i.e. compute a weighted combination of local SPFH signatures) ...
00344       this->weightPointSPFHSignature (hist_f1_, hist_f2_, hist_f3_, nn_indices, nn_dists, fpfh_histogram_);
00345       output.points.row (idx) = fpfh_histogram_;
00346     }
00347   }
00348   else
00349   {
00350     // Iterate over the entire index vector
00351     for (size_t idx = 0; idx < indices_->size (); ++idx)
00352     {
00353       if (this->searchForNeighbors ((*indices_)[idx], search_parameter_, nn_indices, nn_dists) == 0)
00354       {
00355         output.points.row (idx).setConstant (std::numeric_limits<float>::quiet_NaN ());
00356         output.is_dense = false;
00357         continue;
00358       }
00359 
00360       // ... and remap the nn_indices values so that they represent row indices in the spfh_hist_* matrices 
00361       // instead of indices into surface_->points
00362       for (size_t i = 0; i < nn_indices.size (); ++i)
00363         nn_indices[i] = spfh_hist_lookup[nn_indices[i]];
00364 
00365       // Compute the FPFH signature (i.e. compute a weighted combination of local SPFH signatures) ...
00366       this->weightPointSPFHSignature (hist_f1_, hist_f2_, hist_f3_, nn_indices, nn_dists, fpfh_histogram_);
00367       output.points.row (idx) = fpfh_histogram_;
00368     }
00369   }
00370 }
00371 
00372 
00373 #define PCL_INSTANTIATE_FPFHEstimation(T,NT,OutT) template class PCL_EXPORTS pcl::FPFHEstimation<T,NT,OutT>;
00374 
00375 #endif    // PCL_FEATURES_IMPL_FPFH_H_ 
00376 


pcl
Author(s): Open Perception
autogenerated on Mon Oct 6 2014 03:15:12