Go to the documentation of this file.00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021
00022
00023
00024
00025
00026
00027
00028
00029
00030
00031
00032
00033
00034
00035
00036
00037
00038
00039
00040 #ifndef PCL_FEATURES_IMPL_STATISTICAL_MULTISCALE_INTEREST_REGION_EXTRACTION_H_
00041 #define PCL_FEATURES_IMPL_STATISTICAL_MULTISCALE_INTEREST_REGION_EXTRACTION_H_
00042
00043 #include <pcl/features/statistical_multiscale_interest_region_extraction.h>
00044 #include <pcl/kdtree/kdtree_flann.h>
00045 #include <pcl/common/distances.h>
00046 #include <pcl/features/boost.h>
00047 #include <boost/graph/adjacency_list.hpp>
00048 #include <boost/graph/johnson_all_pairs_shortest.hpp>
00049
00050
00052 template <typename PointT> void
00053 pcl::StatisticalMultiscaleInterestRegionExtraction<PointT>::generateCloudGraph ()
00054 {
00055
00056 pcl::KdTreeFLANN<PointT> kdtree;
00057 kdtree.setInputCloud (input_);
00058
00059 using namespace boost;
00060 typedef property<edge_weight_t, float> Weight;
00061 typedef adjacency_list<vecS, vecS, undirectedS, no_property, Weight> Graph;
00062 Graph cloud_graph;
00063
00064 for (size_t point_i = 0; point_i < input_->points.size (); ++point_i)
00065 {
00066 std::vector<int> k_indices (16);
00067 std::vector<float> k_distances (16);
00068 kdtree.nearestKSearch (static_cast<int> (point_i), 16, k_indices, k_distances);
00069
00070 for (int k_i = 0; k_i < static_cast<int> (k_indices.size ()); ++k_i)
00071 add_edge (point_i, k_indices[k_i], Weight (sqrtf (k_distances[k_i])), cloud_graph);
00072 }
00073
00074 const size_t E = num_edges (cloud_graph),
00075 V = num_vertices (cloud_graph);
00076 PCL_INFO ("The graph has %lu vertices and %lu edges.\n", V, E);
00077 geodesic_distances_.clear ();
00078 for (size_t i = 0; i < V; ++i)
00079 {
00080 std::vector<float> aux (V);
00081 geodesic_distances_.push_back (aux);
00082 }
00083 johnson_all_pairs_shortest_paths (cloud_graph, geodesic_distances_);
00084
00085 PCL_INFO ("Done generating the graph\n");
00086 }
00087
00088
00090 template <typename PointT> bool
00091 pcl::StatisticalMultiscaleInterestRegionExtraction<PointT>::initCompute ()
00092 {
00093 if (!PCLBase<PointT>::initCompute ())
00094 {
00095 PCL_ERROR ("[pcl::StatisticalMultiscaleInterestRegionExtraction::initCompute] PCLBase::initCompute () failed - no input cloud was given.\n");
00096 return (false);
00097 }
00098 if (scale_values_.empty ())
00099 {
00100 PCL_ERROR ("[pcl::StatisticalMultiscaleInterestRegionExtraction::initCompute] No scale values were given\n");
00101 return (false);
00102 }
00103
00104 return (true);
00105 }
00106
00107
00109 template <typename PointT> void
00110 pcl::StatisticalMultiscaleInterestRegionExtraction<PointT>::geodesicFixedRadiusSearch (size_t &query_index,
00111 float &radius,
00112 std::vector<int> &result_indices)
00113 {
00114 for (size_t i = 0; i < geodesic_distances_[query_index].size (); ++i)
00115 if (i != query_index && geodesic_distances_[query_index][i] < radius)
00116 result_indices.push_back (static_cast<int> (i));
00117 }
00118
00119
00121 template <typename PointT> void
00122 pcl::StatisticalMultiscaleInterestRegionExtraction<PointT>::computeRegionsOfInterest (std::list<IndicesPtr> &rois)
00123 {
00124 if (!initCompute ())
00125 {
00126 PCL_ERROR ("StatisticalMultiscaleInterestRegionExtraction: not completely initialized\n");
00127 return;
00128 }
00129
00130 generateCloudGraph ();
00131
00132 computeF ();
00133
00134 extractExtrema (rois);
00135 }
00136
00137
00139 template <typename PointT> void
00140 pcl::StatisticalMultiscaleInterestRegionExtraction<PointT>::computeF ()
00141 {
00142 PCL_INFO ("Calculating statistical information\n");
00143
00144
00145 F_scales_.resize (scale_values_.size ());
00146 std::vector<float> point_density (input_->points.size ()),
00147 F (input_->points.size ());
00148 std::vector<std::vector<float> > phi (input_->points.size ());
00149 std::vector<float> phi_row (input_->points.size ());
00150
00151 for (size_t scale_i = 0; scale_i < scale_values_.size (); ++scale_i)
00152 {
00153 float scale_squared = scale_values_[scale_i] * scale_values_[scale_i];
00154
00155
00156 for (size_t point_i = 0; point_i < input_->points.size (); ++point_i)
00157 {
00158 float point_density_i = 0.0;
00159 for (size_t point_j = 0; point_j < input_->points.size (); ++point_j)
00160 {
00161 float d_g = geodesic_distances_[point_i][point_j];
00162 float phi_i_j = 1.0f / sqrtf (2.0f * static_cast<float> (M_PI) * scale_squared) * expf ( (-1) * d_g*d_g / (2.0f * scale_squared));
00163
00164 point_density_i += phi_i_j;
00165 phi_row[point_j] = phi_i_j;
00166 }
00167 point_density[point_i] = point_density_i;
00168 phi[point_i] = phi_row;
00169 }
00170
00171
00172 for (size_t point_i = 0; point_i < input_->points.size (); ++point_i)
00173 {
00174 float A_hat_normalization = 0.0;
00175 PointT A_hat; A_hat.x = A_hat.y = A_hat.z = 0.0;
00176 for (size_t point_j = 0; point_j < input_->points.size (); ++point_j)
00177 {
00178 float phi_hat_i_j = phi[point_i][point_j] / (point_density[point_i] * point_density[point_j]);
00179 A_hat_normalization += phi_hat_i_j;
00180
00181 PointT aux = input_->points[point_j];
00182 aux.x *= phi_hat_i_j; aux.y *= phi_hat_i_j; aux.z *= phi_hat_i_j;
00183
00184 A_hat.x += aux.x; A_hat.y += aux.y; A_hat.z += aux.z;
00185 }
00186 A_hat.x /= A_hat_normalization; A_hat.y /= A_hat_normalization; A_hat.z /= A_hat_normalization;
00187
00188
00189 float aux = 2.0f / scale_values_[scale_i] * euclideanDistance<PointT, PointT> (A_hat, input_->points[point_i]);
00190 F[point_i] = aux * expf (-aux);
00191 }
00192
00193 F_scales_[scale_i] = F;
00194 }
00195 }
00196
00197
00199 template <typename PointT> void
00200 pcl::StatisticalMultiscaleInterestRegionExtraction<PointT>::extractExtrema (std::list<IndicesPtr> &rois)
00201 {
00202 std::vector<std::vector<bool> > is_min (scale_values_.size ()),
00203 is_max (scale_values_.size ());
00204
00205
00206 for (size_t scale_i = 0; scale_i < scale_values_.size (); ++scale_i)
00207 {
00208 std::vector<bool> is_min_scale (input_->points.size ()),
00209 is_max_scale (input_->points.size ());
00210 for (size_t point_i = 0; point_i < input_->points.size (); ++point_i)
00211 {
00212 std::vector<int> nn_indices;
00213 geodesicFixedRadiusSearch (point_i, scale_values_[scale_i], nn_indices);
00214 bool is_max_point = true, is_min_point = true;
00215 for (std::vector<int>::iterator nn_it = nn_indices.begin (); nn_it != nn_indices.end (); ++nn_it)
00216 if (F_scales_[scale_i][point_i] < F_scales_[scale_i][*nn_it])
00217 is_max_point = false;
00218 else
00219 is_min_point = false;
00220
00221 is_min_scale[point_i] = is_min_point;
00222 is_max_scale[point_i] = is_max_point;
00223 }
00224
00225 is_min[scale_i] = is_min_scale;
00226 is_max[scale_i] = is_max_scale;
00227 }
00228
00229
00230 for (size_t scale_i = 1; scale_i < scale_values_.size () - 1; ++scale_i)
00231 {
00232 for (size_t point_i = 0; point_i < input_->points.size (); ++point_i)
00233 if ((is_min[scale_i - 1][point_i] && is_min[scale_i][point_i] && is_min[scale_i + 1][point_i]) ||
00234 (is_max[scale_i - 1][point_i] && is_max[scale_i][point_i] && is_max[scale_i + 1][point_i]))
00235 {
00236
00237 IndicesPtr region (new std::vector<int>);
00238 region->push_back (static_cast<int> (point_i));
00239
00240
00241 std::vector<int> nn_indices;
00242 geodesicFixedRadiusSearch (point_i, scale_values_[scale_i], nn_indices);
00243 region->insert (region->end (), nn_indices.begin (), nn_indices.end ());
00244 rois.push_back (region);
00245 }
00246 }
00247 }
00248
00249
00250 #define PCL_INSTANTIATE_StatisticalMultiscaleInterestRegionExtraction(T) template class PCL_EXPORTS pcl::StatisticalMultiscaleInterestRegionExtraction<T>;
00251
00252 #endif
00253