Point Cloud Library (PCL) 1.13.1
Loading...
Searching...
No Matches
sift_keypoint.hpp
1/*
2 * Software License Agreement (BSD License)
3 *
4 * Point Cloud Library (PCL) - www.pointclouds.org
5 * Copyright (c) 2010-2011, 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 Willow Garage, Inc. 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#ifndef PCL_SIFT_KEYPOINT_IMPL_H_
39#define PCL_SIFT_KEYPOINT_IMPL_H_
40
41#include <pcl/keypoints/sift_keypoint.h>
42#include <pcl/common/io.h>
43#include <pcl/filters/voxel_grid.h>
44
45//////////////////////////////////////////////////////////////////////////////////////////////////////////////////
46template <typename PointInT, typename PointOutT> void
47pcl::SIFTKeypoint<PointInT, PointOutT>::setScales (float min_scale, int nr_octaves, int nr_scales_per_octave)
48{
49 min_scale_ = min_scale;
50 nr_octaves_ = nr_octaves;
51 nr_scales_per_octave_ = nr_scales_per_octave;
52}
53
54
55//////////////////////////////////////////////////////////////////////////////////////////////////////////////////
56template <typename PointInT, typename PointOutT> void
58{
59 min_contrast_ = min_contrast;
60}
61
62//////////////////////////////////////////////////////////////////////////////////////////////////////////////////
63template <typename PointInT, typename PointOutT> bool
65{
66 if (min_scale_ <= 0)
67 {
68 PCL_ERROR ("[pcl::%s::initCompute] : Minimum scale (%f) must be strict positive!\n",
69 name_.c_str (), min_scale_);
70 return (false);
71 }
72 if (nr_octaves_ < 1)
73 {
74 PCL_ERROR ("[pcl::%s::initCompute] : Number of octaves (%d) must be at least 1!\n",
75 name_.c_str (), nr_octaves_);
76 return (false);
77 }
78 if (nr_scales_per_octave_ < 1)
79 {
80 PCL_ERROR ("[pcl::%s::initCompute] : Number of scales per octave (%d) must be at least 1!\n",
81 name_.c_str (), nr_scales_per_octave_);
82 return (false);
83 }
84 if (min_contrast_ < 0)
85 {
86 PCL_ERROR ("[pcl::%s::initCompute] : Minimum contrast (%f) must be non-negative!\n",
87 name_.c_str (), min_contrast_);
88 return (false);
89 }
90
91 this->setKSearch (1);
92 tree_.reset (new pcl::search::KdTree<PointInT> (true));
93 return (true);
94}
95
96//////////////////////////////////////////////////////////////////////////////////////////////////////////////////
97template <typename PointInT, typename PointOutT> void
99{
100 if (surface_ && surface_ != input_)
101 {
102 PCL_WARN ("[pcl::%s::detectKeypoints] : ", name_.c_str ());
103 PCL_WARN ("A search surface has been set by setSearchSurface, but this SIFT keypoint detection algorithm does ");
104 PCL_WARN ("not support search surfaces other than the input cloud. ");
105 PCL_WARN ("The cloud provided in setInputCloud is being used instead.\n");
106 }
107
108 // Check if the output has a "scale" field
109 scale_idx_ = pcl::getFieldIndex<PointOutT> ("scale", out_fields_);
110
111 // Make sure the output cloud is empty
112 output.clear ();
113
114 // Create a local copy of the input cloud that will be resized for each octave
115 typename pcl::PointCloud<PointInT>::Ptr cloud (new pcl::PointCloud<PointInT> (*input_));
116
117 VoxelGrid<PointInT> voxel_grid;
118 // Search for keypoints at each octave
119 float scale = min_scale_;
120 for (int i_octave = 0; i_octave < nr_octaves_; ++i_octave)
121 {
122 // Downsample the point cloud
123 const float s = 1.0f * scale; // note: this can be adjusted
124 voxel_grid.setLeafSize (s, s, s);
125 voxel_grid.setInputCloud (cloud);
127 voxel_grid.filter (*temp);
128 cloud = temp;
129
130 // Make sure the downsampled cloud still has enough points
131 constexpr std::size_t min_nr_points = 25;
132 if (cloud->size () < min_nr_points)
133 break;
134
135 // Update the KdTree with the downsampled points
136 tree_->setInputCloud (cloud);
137
138 // Detect keypoints for the current scale
139 detectKeypointsForOctave (*cloud, *tree_, scale, nr_scales_per_octave_, output);
140
141 // Increase the scale by another octave
142 scale *= 2;
143 }
144
145 // Set final properties
146 output.height = 1;
147 output.width = output.size ();
148 output.header = input_->header;
149 output.sensor_origin_ = input_->sensor_origin_;
150 output.sensor_orientation_ = input_->sensor_orientation_;
151}
152
153
154//////////////////////////////////////////////////////////////////////////////////////////////////////////////////
155template <typename PointInT, typename PointOutT> void
157 const PointCloudIn &input, KdTree &tree, float base_scale, int nr_scales_per_octave,
158 PointCloudOut &output)
159{
160 // Compute the difference of Gaussians (DoG) scale space
161 std::vector<float> scales (nr_scales_per_octave + 3);
162 for (int i_scale = 0; i_scale <= nr_scales_per_octave + 2; ++i_scale)
163 {
164 scales[i_scale] = base_scale * powf (2.0f, (1.0f * static_cast<float> (i_scale) - 1.0f) / static_cast<float> (nr_scales_per_octave));
165 }
166 Eigen::MatrixXf diff_of_gauss;
167 computeScaleSpace (input, tree, scales, diff_of_gauss);
168
169 // Find extrema in the DoG scale space
170 pcl::Indices extrema_indices;
171 std::vector<int> extrema_scales;
172 findScaleSpaceExtrema (input, tree, diff_of_gauss, extrema_indices, extrema_scales);
173
174 output.reserve (output.size () + extrema_indices.size ());
175 // Save scale?
176 if (scale_idx_ != -1)
177 {
178 // Add keypoints to output
179 for (std::size_t i_keypoint = 0; i_keypoint < extrema_indices.size (); ++i_keypoint)
180 {
181 PointOutT keypoint;
182 const auto &keypoint_index = extrema_indices[i_keypoint];
183
184 keypoint.x = input[keypoint_index].x;
185 keypoint.y = input[keypoint_index].y;
186 keypoint.z = input[keypoint_index].z;
187 memcpy (reinterpret_cast<char*> (&keypoint) + out_fields_[scale_idx_].offset,
188 &scales[extrema_scales[i_keypoint]], sizeof (float));
189 output.push_back (keypoint);
190 }
191 }
192 else
193 {
194 // Add keypoints to output
195 for (const auto &keypoint_index : extrema_indices)
196 {
197 PointOutT keypoint;
198 keypoint.x = input[keypoint_index].x;
199 keypoint.y = input[keypoint_index].y;
200 keypoint.z = input[keypoint_index].z;
201
202 output.push_back (keypoint);
203 }
204 }
205}
206
207
208//////////////////////////////////////////////////////////////////////////////////////////////////////////////////
209template <typename PointInT, typename PointOutT>
211 const PointCloudIn &input, KdTree &tree, const std::vector<float> &scales,
212 Eigen::MatrixXf &diff_of_gauss)
213{
214 diff_of_gauss.resize (input.size (), scales.size () - 1);
215
216 // For efficiency, we will only filter over points within 3 standard deviations
217 const float max_radius = 3.0f * scales.back ();
218
219 for (int i_point = 0; i_point < static_cast<int> (input.size ()); ++i_point)
220 {
221 pcl::Indices nn_indices;
222 std::vector<float> nn_dist;
223 tree.radiusSearch (i_point, max_radius, nn_indices, nn_dist); // *
224 // * note: at this stage of the algorithm, we must find all points within a radius defined by the maximum scale,
225 // regardless of the configurable search method specified by the user, so we directly employ tree.radiusSearch
226 // here instead of using searchForNeighbors.
227
228 // For each scale, compute the Gaussian "filter response" at the current point
229 float filter_response = 0.0f;
230 for (std::size_t i_scale = 0; i_scale < scales.size (); ++i_scale)
231 {
232 float sigma_sqr = powf (scales[i_scale], 2.0f);
233
234 float numerator = 0.0f;
235 float denominator = 0.0f;
236 for (std::size_t i_neighbor = 0; i_neighbor < nn_indices.size (); ++i_neighbor)
237 {
238 const float &value = getFieldValue_ (input[nn_indices[i_neighbor]]);
239 const float &dist_sqr = nn_dist[i_neighbor];
240 if (dist_sqr <= 9*sigma_sqr)
241 {
242 float w = std::exp (-0.5f * dist_sqr / sigma_sqr);
243 numerator += value * w;
244 denominator += w;
245 }
246 else break; // i.e. if dist > 3 standard deviations, then terminate early
247 }
248 float previous_filter_response = filter_response;
249 filter_response = numerator / denominator;
250
251 // Compute the difference between adjacent scales
252 if (i_scale > 0)
253 diff_of_gauss (i_point, i_scale - 1) = filter_response - previous_filter_response;
254 }
255 }
256}
257
258//////////////////////////////////////////////////////////////////////////////////////////////////////////////////
259template <typename PointInT, typename PointOutT> void
261 const PointCloudIn &input, KdTree &tree, const Eigen::MatrixXf &diff_of_gauss,
262 pcl::Indices &extrema_indices, std::vector<int> &extrema_scales)
263{
264 constexpr int k = 25;
265 pcl::Indices nn_indices (k);
266 std::vector<float> nn_dist (k);
267
268 const int nr_scales = static_cast<int> (diff_of_gauss.cols ());
269 std::vector<float> min_val (nr_scales), max_val (nr_scales);
270
271 for (int i_point = 0; i_point < static_cast<int> (input.size ()); ++i_point)
272 {
273 // Define the local neighborhood around the current point
274 const std::size_t nr_nn = tree.nearestKSearch (i_point, k, nn_indices, nn_dist); //*
275 // * note: the neighborhood for finding local extrema is best defined as a small fixed-k neighborhood, regardless of
276 // the configurable search method specified by the user, so we directly employ tree.nearestKSearch here instead
277 // of using searchForNeighbors
278
279 // At each scale, find the extreme values of the DoG within the current neighborhood
280 for (int i_scale = 0; i_scale < nr_scales; ++i_scale)
281 {
282 min_val[i_scale] = std::numeric_limits<float>::max ();
283 max_val[i_scale] = -std::numeric_limits<float>::max ();
284
285 for (std::size_t i_neighbor = 0; i_neighbor < nr_nn; ++i_neighbor)
286 {
287 const float &d = diff_of_gauss (nn_indices[i_neighbor], i_scale);
288
289 min_val[i_scale] = (std::min) (min_val[i_scale], d);
290 max_val[i_scale] = (std::max) (max_val[i_scale], d);
291 }
292 }
293
294 // If the current point is an extreme value with high enough contrast, add it as a keypoint
295 for (int i_scale = 1; i_scale < nr_scales - 1; ++i_scale)
296 {
297 const float &val = diff_of_gauss (i_point, i_scale);
298
299 // Does the point have sufficient contrast?
300 if (std::abs (val) >= min_contrast_)
301 {
302 // Is it a local minimum?
303 if ((val == min_val[i_scale]) &&
304 (val < min_val[i_scale - 1]) &&
305 (val < min_val[i_scale + 1]))
306 {
307 extrema_indices.push_back (i_point);
308 extrema_scales.push_back (i_scale);
309 }
310 // Is it a local maximum?
311 else if ((val == max_val[i_scale]) &&
312 (val > max_val[i_scale - 1]) &&
313 (val > max_val[i_scale + 1]))
314 {
315 extrema_indices.push_back (i_point);
316 extrema_scales.push_back (i_scale);
317 }
318 }
319 }
320 }
321}
322
323#define PCL_INSTANTIATE_SIFTKeypoint(T,U) template class PCL_EXPORTS pcl::SIFTKeypoint<T,U>;
324
325#endif // #ifndef PCL_SIFT_KEYPOINT_IMPL_H_
326
void filter(PointCloud &output)
Calls the filtering method and returns the filtered dataset in output.
Definition filter.h:121
KdTree represents the base spatial locator class for kd-tree implementations.
Definition kdtree.h:55
virtual void setInputCloud(const PointCloudConstPtr &cloud)
Provide a pointer to the input dataset.
Definition pcl_base.hpp:65
PointCloud represents the base class in PCL for storing collections of 3D points.
std::size_t size() const
shared_ptr< PointCloud< PointT > > Ptr
SIFTKeypoint detects the Scale Invariant Feature Transform keypoints for a given point cloud dataset ...
bool initCompute() override
void setMinimumContrast(float min_contrast)
Provide a threshold to limit detection of keypoints without sufficient contrast.
void setScales(float min_scale, int nr_octaves, int nr_scales_per_octave)
Specify the range of scales over which to search for keypoints.
void detectKeypoints(PointCloudOut &output) override
Detect the SIFT keypoints for a set of points given in setInputCloud () using the spatial locator in ...
typename Keypoint< PointInT, PointOutT >::PointCloudOut PointCloudOut
VoxelGrid assembles a local 3D grid over a given PointCloud, and downsamples + filters the data.
Definition voxel_grid.h:177
void setLeafSize(const Eigen::Vector4f &leaf_size)
Set the voxel grid leaf size.
Definition voxel_grid.h:221
search::KdTree is a wrapper class which inherits the pcl::KdTree class for performing search function...
Definition kdtree.h:62
IndicesAllocator<> Indices
Type used for indices in PCL.
Definition types.h:133