Point Cloud Library (PCL) 1.12.0
object_recognition.h
1#pragma once
2
3#include "typedefs.h"
4#include "load_clouds.h"
5#include "solution/filters.h"
6#include "solution/segmentation.h"
7#include "solution/feature_estimation.h"
8#include "solution/registration.h"
9
10#include <pcl/io/pcd_io.h>
11#include <pcl/kdtree/kdtree_flann.h>
12
14{
15 // Filter parameters
16 float min_depth;
17 float max_depth;
21
22 // Segmentation parameters
28
29 // Feature estimation parameters
36
37 // Registration parameters
45};
46
47struct ObjectModel
48{
49 PointCloudPtr points;
50 PointCloudPtr keypoints;
51 LocalDescriptorsPtr local_descriptors;
52 GlobalDescriptorsPtr global_descriptor;
53};
54
56{
57 public:
59 {}
60
61 void
62 populateDatabase (const std::vector<std::string> & filenames)
63 {
64 std::size_t n = filenames.size ();
65 models_.resize (n);
66 descriptors_ = GlobalDescriptorsPtr (new GlobalDescriptors);
67 for (std::size_t i = 0; i < n; ++i)
68 {
69 const std::string & filename = filenames[i];
70 if (filename.compare (filename.size ()-4, 4, ".pcd") == 0)
71 {
72 // If filename ends pcd extension, load the points and process them
73 PointCloudPtr raw_input (new PointCloud);
74 pcl::io::loadPCDFile (filenames[i], *raw_input);
75
76 constructObjectModel (raw_input, models_[i]);
77 }
78 else
79 {
80 // If the filename has no extension, load the pre-computed models
81 models_[i].points = loadPointCloud<PointT> (filename, "_points.pcd");
82 models_[i].keypoints = loadPointCloud<PointT> (filename, "_keypoints.pcd");
83 models_[i].local_descriptors = loadPointCloud<LocalDescriptorT> (filename, "_localdesc.pcd");
84 models_[i].global_descriptor = loadPointCloud<GlobalDescriptorT> (filename, "_globaldesc.pcd");
85 }
86 *descriptors_ += *(models_[i].global_descriptor);
87 }
89 kdtree_->setInputCloud (descriptors_);
90 }
91
92 const ObjectModel &
93 recognizeObject (const PointCloudPtr & query_cloud)
94 {
95 ObjectModel query_object;
96 constructObjectModel (query_cloud, query_object);
97 const GlobalDescriptorT & query_descriptor = (*query_object.global_descriptor)[0];
98
99 std::vector<int> nn_index (1);
100 std::vector<float> nn_sqr_distance (1);
101 kdtree_->nearestKSearch (query_descriptor, 1, nn_index, nn_sqr_distance);
102 const int & best_match = nn_index[0];
103
104 return (models_[best_match]);
105 }
106
107 PointCloudPtr
108 recognizeAndAlignPoints (const PointCloudPtr & query_cloud)
109 {
110 ObjectModel query_object;
111 constructObjectModel (query_cloud, query_object);
112 const GlobalDescriptorT & query_descriptor = (*query_object.global_descriptor)[0];
113
114 std::vector<int> nn_index (1);
115 std::vector<float> nn_sqr_distance (1);
116 kdtree_->nearestKSearch (query_descriptor, 1, nn_index, nn_sqr_distance);
117 const int & best_match = nn_index[0];
118
119 PointCloudPtr output = alignModelPoints (models_[best_match], query_object, params_);
120 return (output);
121 }
122
123 /* Construct an object model by filtering, segmenting, and estimating feature descriptors */
124 void
125 constructObjectModel (const PointCloudPtr & points, ObjectModel & output) const
126 {
127 output.points = applyFiltersAndSegment (points, params_);
128
129 SurfaceNormalsPtr normals;
130 estimateFeatures (output.points, params_, normals, output.keypoints,
131 output.local_descriptors, output.global_descriptor);
132 }
133
134 protected:
135 /* Apply a series of filters (threshold depth, downsample, and remove outliers) */
136 PointCloudPtr
137 applyFiltersAndSegment (const PointCloudPtr & input, const ObjectRecognitionParameters & params) const
138 {
139 PointCloudPtr cloud;
140 cloud = thresholdDepth (input, params.min_depth, params.max_depth);
141 cloud = downsample (cloud, params.downsample_leaf_size);
142 cloud = removeOutliers (cloud, params.outlier_rejection_radius, params.outlier_rejection_min_neighbors);
143
144 cloud = findAndSubtractPlane (cloud, params.plane_inlier_distance_threshold, params.max_ransac_iterations);
145 std::vector<pcl::PointIndices> cluster_indices;
146 clusterObjects (cloud, params.cluster_tolerance, params.min_cluster_size,
147 params.max_cluster_size, cluster_indices);
148
149 PointCloudPtr largest_cluster (new PointCloud);
150 pcl::copyPointCloud (*cloud, cluster_indices[0], *largest_cluster);
151
152 return (largest_cluster);
153 }
154
155 /* Estimate surface normals, keypoints, and local/global feature descriptors */
156 void
157 estimateFeatures (const PointCloudPtr & points, const ObjectRecognitionParameters & params,
158 SurfaceNormalsPtr & normals_out, PointCloudPtr & keypoints_out,
159 LocalDescriptorsPtr & local_descriptors_out, GlobalDescriptorsPtr & global_descriptor_out) const
160 {
161 normals_out = estimateSurfaceNormals (points, params.surface_normal_radius);
162
163 keypoints_out = detectKeypoints (points, normals_out, params.keypoints_min_scale, params.keypoints_nr_octaves,
165
166 local_descriptors_out = computeLocalDescriptors (points, normals_out, keypoints_out,
168
169 global_descriptor_out = computeGlobalDescriptor (points, normals_out);
170 }
171
172 /* Align the points in the source model to the points in the target model */
173 PointCloudPtr
174 alignModelPoints (const ObjectModel & source, const ObjectModel & target,
175 const ObjectRecognitionParameters & params) const
176 {
177 Eigen::Matrix4f tform;
178 tform = computeInitialAlignment (source.keypoints, source.local_descriptors,
179 target.keypoints, target.local_descriptors,
183
184 tform = refineAlignment (source.points, target.points, tform,
187
188 PointCloudPtr output (new PointCloud);
189 pcl::transformPointCloud (*(source.points), *output, tform);
190
191 return (output);
192 }
193
195 std::vector<ObjectModel> models_;
196 GlobalDescriptorsPtr descriptors_;
198};
pcl::KdTreeFLANN< GlobalDescriptorT >::Ptr kdtree_
PointCloudPtr applyFiltersAndSegment(const PointCloudPtr &input, const ObjectRecognitionParameters &params) const
std::vector< ObjectModel > models_
const ObjectModel & recognizeObject(const PointCloudPtr &query_cloud)
GlobalDescriptorsPtr descriptors_
PointCloudPtr recognizeAndAlignPoints(const PointCloudPtr &query_cloud)
void populateDatabase(const std::vector< std::string > &filenames)
ObjectRecognition(const ObjectRecognitionParameters &params)
ObjectRecognitionParameters params_
PointCloudPtr alignModelPoints(const ObjectModel &source, const ObjectModel &target, const ObjectRecognitionParameters &params) const
void constructObjectModel(const PointCloudPtr &points, ObjectModel &output) const
void estimateFeatures(const PointCloudPtr &points, const ObjectRecognitionParameters &params, SurfaceNormalsPtr &normals_out, PointCloudPtr &keypoints_out, LocalDescriptorsPtr &local_descriptors_out, GlobalDescriptorsPtr &global_descriptor_out) const
KdTreeFLANN is a generic type of 3D spatial locator using kD-tree structures.
Definition: kdtree_flann.h:132
shared_ptr< KdTreeFLANN< PointT, Dist > > Ptr
Definition: kdtree_flann.h:151
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
void copyPointCloud(const pcl::PointCloud< PointInT > &cloud_in, pcl::PointCloud< PointOutT > &cloud_out)
Copy all the fields from a given point cloud into a new point cloud.
Definition: io.hpp:144
int loadPCDFile(const std::string &file_name, pcl::PCLPointCloud2 &cloud)
Load a PCD v.6 file into a templated PointCloud type.
Definition: pcd_io.h:621
GlobalDescriptorsPtr global_descriptor
LocalDescriptorsPtr local_descriptors
PointCloudPtr keypoints
PointCloudPtr points
A point structure representing the Viewpoint Feature Histogram (VFH).