Point Cloud Library (PCL)
1.3.1
|
00001 /* 00002 * Software License Agreement (BSD License) 00003 * 00004 * Copyright (c) 2009, Willow Garage, Inc. 00005 * All rights reserved. 00006 * 00007 * Redistribution and use in source and binary forms, with or without 00008 * modification, are permitted provided that the following conditions 00009 * are met: 00010 * 00011 * * Redistributions of source code must retain the above copyright 00012 * notice, this list of conditions and the following disclaimer. 00013 * * Redistributions in binary form must reproduce the above 00014 * copyright notice, this list of conditions and the following 00015 * disclaimer in the documentation and/or other materials provided 00016 * with the distribution. 00017 * * Neither the name of Willow Garage, Inc. nor the names of its 00018 * contributors may be used to endorse or promote products derived 00019 * from this software without specific prior written permission. 00020 * 00021 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 00022 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 00023 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 00024 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 00025 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 00026 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 00027 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 00028 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 00029 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 00030 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 00031 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 00032 * POSSIBILITY OF SUCH DAMAGE. 00033 * 00034 * $Id: mlesac.hpp 3280 2011-11-30 18:31:35Z gedikli $ 00035 * 00036 */ 00037 00038 #ifndef PCL_SAMPLE_CONSENSUS_IMPL_MLESAC_H_ 00039 #define PCL_SAMPLE_CONSENSUS_IMPL_MLESAC_H_ 00040 00041 #include "pcl/sample_consensus/mlesac.h" 00042 00044 template <typename PointT> bool 00045 pcl::MaximumLikelihoodSampleConsensus<PointT>::computeModel (int debug_verbosity_level) 00046 { 00047 // Warn and exit if no threshold was set 00048 if (threshold_ == DBL_MAX) 00049 { 00050 PCL_ERROR ("[pcl::MaximumLikelihoodSampleConsensus::computeModel] No threshold set!\n"); 00051 return (false); 00052 } 00053 00054 iterations_ = 0; 00055 double d_best_penalty = DBL_MAX; 00056 double k = 1.0; 00057 00058 std::vector<int> best_model; 00059 std::vector<int> selection; 00060 Eigen::VectorXf model_coefficients; 00061 std::vector<double> distances; 00062 00063 // Compute sigma - remember to set threshold_ correctly ! 00064 sigma_ = computeMedianAbsoluteDeviation (sac_model_->getInputCloud (), sac_model_->getIndices (), threshold_); 00065 if (debug_verbosity_level > 1) 00066 PCL_DEBUG ("[pcl::MaximumLikelihoodSampleConsensus::computeModel] Estimated sigma value: %f.\n", sigma_); 00067 00068 // Compute the bounding box diagonal: V = sqrt (sum (max(pointCloud) - min(pointCloud)^2)) 00069 Eigen::Vector4f min_pt, max_pt; 00070 getMinMax (sac_model_->getInputCloud (), sac_model_->getIndices (), min_pt, max_pt); 00071 max_pt -= min_pt; 00072 double v = sqrt (max_pt.dot (max_pt)); 00073 00074 int n_inliers_count = 0; 00075 size_t indices_size; 00076 unsigned skipped_count = 0; 00077 // supress infinite loops by just allowing 10 x maximum allowed iterations for invalid model parameters! 00078 const unsigned max_skip = max_iterations_ * 10; 00079 00080 // Iterate 00081 while (iterations_ < k && skipped_count < max_skip) 00082 { 00083 // Get X samples which satisfy the model criteria 00084 sac_model_->getSamples (iterations_, selection); 00085 00086 if (selection.empty ()) break; 00087 00088 // Search for inliers in the point cloud for the current plane model M 00089 if (!sac_model_->computeModelCoefficients (selection, model_coefficients)) 00090 { 00091 //iterations_++; 00092 ++ skipped_count; 00093 continue; 00094 } 00095 00096 // Iterate through the 3d points and calculate the distances from them to the model 00097 sac_model_->getDistancesToModel (model_coefficients, distances); 00098 00099 // Use Expectiation-Maximization to find out the right value for d_cur_penalty 00100 // ---[ Initial estimate for the gamma mixing parameter = 1/2 00101 double gamma = 0.5; 00102 double p_outlier_prob = 0; 00103 00104 indices_size = sac_model_->getIndices ()->size (); 00105 std::vector<double> p_inlier_prob (indices_size); 00106 for (int j = 0; j < iterations_EM_; ++j) 00107 { 00108 // Likelihood of a datum given that it is an inlier 00109 for (size_t i = 0; i < indices_size; ++i) 00110 p_inlier_prob[i] = gamma * exp (- (distances[i] * distances[i] ) / 2 * (sigma_ * sigma_) ) / 00111 (sqrt (2 * M_PI) * sigma_); 00112 00113 // Likelihood of a datum given that it is an outlier 00114 p_outlier_prob = (1 - gamma) / v; 00115 00116 gamma = 0; 00117 for (size_t i = 0; i < indices_size; ++i) 00118 gamma += p_inlier_prob [i] / (p_inlier_prob[i] + p_outlier_prob); 00119 gamma /= sac_model_->getIndices ()->size (); 00120 } 00121 00122 // Find the log likelihood of the model -L = -sum [log (pInlierProb + pOutlierProb)] 00123 double d_cur_penalty = 0; 00124 for (size_t i = 0; i < indices_size; ++i) 00125 d_cur_penalty += log (p_inlier_prob[i] + p_outlier_prob); 00126 d_cur_penalty = - d_cur_penalty; 00127 00128 // Better match ? 00129 if (d_cur_penalty < d_best_penalty) 00130 { 00131 d_best_penalty = d_cur_penalty; 00132 00133 // Save the current model/coefficients selection as being the best so far 00134 model_ = selection; 00135 model_coefficients_ = model_coefficients; 00136 00137 n_inliers_count = 0; 00138 // Need to compute the number of inliers for this model to adapt k 00139 for (size_t i = 0; i < distances.size (); ++i) 00140 if (distances[i] <= 2 * sigma_) 00141 n_inliers_count++; 00142 00143 // Compute the k parameter (k=log(z)/log(1-w^n)) 00144 double w = (double)((double)n_inliers_count / (double)sac_model_->getIndices ()->size ()); 00145 double p_no_outliers = 1 - pow (w, (double)selection.size ()); 00146 p_no_outliers = (std::max) (std::numeric_limits<double>::epsilon (), p_no_outliers); // Avoid division by -Inf 00147 p_no_outliers = (std::min) (1 - std::numeric_limits<double>::epsilon (), p_no_outliers); // Avoid division by 0. 00148 k = log (1 - probability_) / log (p_no_outliers); 00149 } 00150 00151 ++iterations_; 00152 if (debug_verbosity_level > 1) 00153 PCL_DEBUG ("[pcl::MaximumLikelihoodSampleConsensus::computeModel] Trial %d out of %d. Best penalty is %f.\n", iterations_, (int)ceil (k), d_best_penalty); 00154 if (iterations_ > max_iterations_) 00155 { 00156 if (debug_verbosity_level > 0) 00157 PCL_DEBUG ("[pcl::MaximumLikelihoodSampleConsensus::computeModel] MLESAC reached the maximum number of trials.\n"); 00158 break; 00159 } 00160 } 00161 00162 if (model_.empty ()) 00163 { 00164 if (debug_verbosity_level > 0) 00165 PCL_DEBUG ("[pcl::MaximumLikelihoodSampleConsensus::computeModel] Unable to find a solution!\n"); 00166 return (false); 00167 } 00168 00169 // Iterate through the 3d points and calculate the distances from them to the model again 00170 sac_model_->getDistancesToModel (model_coefficients_, distances); 00171 std::vector<int> &indices = *sac_model_->getIndices (); 00172 if (distances.size () != indices.size ()) 00173 { 00174 PCL_ERROR ("[pcl::MaximumLikelihoodSampleConsensus::computeModel] Estimated distances (%lu) differs than the normal of indices (%lu).\n", (unsigned long)distances.size (), (unsigned long)indices.size ()); 00175 return (false); 00176 } 00177 00178 inliers_.resize (distances.size ()); 00179 // Get the inliers for the best model found 00180 n_inliers_count = 0; 00181 for (size_t i = 0; i < distances.size (); ++i) 00182 if (distances[i] <= 2 * sigma_) 00183 inliers_[n_inliers_count++] = indices[i]; 00184 00185 // Resize the inliers vector 00186 inliers_.resize (n_inliers_count); 00187 00188 if (debug_verbosity_level > 0) 00189 PCL_DEBUG ("[pcl::MaximumLikelihoodSampleConsensus::computeModel] Model: %lu size, %d inliers.\n", (unsigned long)model_.size (), n_inliers_count); 00190 00191 return (true); 00192 } 00193 00195 template <typename PointT> double 00196 pcl::MaximumLikelihoodSampleConsensus<PointT>::computeMedianAbsoluteDeviation ( 00197 const PointCloudConstPtr &cloud, 00198 const boost::shared_ptr <std::vector<int> > &indices, 00199 double sigma) 00200 { 00201 std::vector<double> distances (indices->size ()); 00202 00203 Eigen::Vector4f median; 00204 // median (dist (x - median (x))) 00205 computeMedian (cloud, indices, median); 00206 00207 for (size_t i = 0; i < indices->size (); ++i) 00208 { 00209 pcl::Vector4fMapConst pt = cloud->points[(*indices)[i]].getVector4fMap (); 00210 Eigen::Vector4f ptdiff = pt - median; 00211 ptdiff[3] = 0; 00212 distances[i] = ptdiff.dot (ptdiff); 00213 } 00214 00215 std::sort (distances.begin (), distances.end ()); 00216 00217 double result; 00218 int mid = indices->size () / 2; 00219 // Do we have a "middle" point or should we "estimate" one ? 00220 if (indices->size () % 2 == 0) 00221 result = (sqrt (distances[mid-1]) + sqrt (distances[mid])) / 2; 00222 else 00223 result = sqrt (distances[mid]); 00224 return (sigma * result); 00225 } 00226 00228 template <typename PointT> void 00229 pcl::MaximumLikelihoodSampleConsensus<PointT>::getMinMax ( 00230 const PointCloudConstPtr &cloud, 00231 const boost::shared_ptr <std::vector<int> > &indices, 00232 Eigen::Vector4f &min_p, 00233 Eigen::Vector4f &max_p) 00234 { 00235 min_p.setConstant (FLT_MAX); 00236 max_p.setConstant (-FLT_MAX); 00237 min_p[3] = max_p[3] = 0; 00238 00239 for (size_t i = 0; i < indices->size (); ++i) 00240 { 00241 if (cloud->points[(*indices)[i]].x < min_p[0]) min_p[0] = cloud->points[(*indices)[i]].x; 00242 if (cloud->points[(*indices)[i]].y < min_p[1]) min_p[1] = cloud->points[(*indices)[i]].y; 00243 if (cloud->points[(*indices)[i]].z < min_p[2]) min_p[2] = cloud->points[(*indices)[i]].z; 00244 00245 if (cloud->points[(*indices)[i]].x > max_p[0]) max_p[0] = cloud->points[(*indices)[i]].x; 00246 if (cloud->points[(*indices)[i]].y > max_p[1]) max_p[1] = cloud->points[(*indices)[i]].y; 00247 if (cloud->points[(*indices)[i]].z > max_p[2]) max_p[2] = cloud->points[(*indices)[i]].z; 00248 } 00249 } 00250 00252 template <typename PointT> void 00253 pcl::MaximumLikelihoodSampleConsensus<PointT>::computeMedian ( 00254 const PointCloudConstPtr &cloud, 00255 const boost::shared_ptr <std::vector<int> > &indices, 00256 Eigen::Vector4f &median) 00257 { 00258 // Copy the values to vectors for faster sorting 00259 std::vector<float> x (indices->size ()); 00260 std::vector<float> y (indices->size ()); 00261 std::vector<float> z (indices->size ()); 00262 for (size_t i = 0; i < indices->size (); ++i) 00263 { 00264 x[i] = cloud->points[(*indices)[i]].x; 00265 y[i] = cloud->points[(*indices)[i]].y; 00266 z[i] = cloud->points[(*indices)[i]].z; 00267 } 00268 std::sort (x.begin (), x.end ()); 00269 std::sort (y.begin (), y.end ()); 00270 std::sort (z.begin (), z.end ()); 00271 00272 int mid = indices->size () / 2; 00273 if (indices->size () % 2 == 0) 00274 { 00275 median[0] = (x[mid-1] + x[mid]) / 2; 00276 median[1] = (y[mid-1] + y[mid]) / 2; 00277 median[2] = (z[mid-1] + z[mid]) / 2; 00278 } 00279 else 00280 { 00281 median[0] = x[mid]; 00282 median[1] = y[mid]; 00283 median[2] = z[mid]; 00284 } 00285 median[3] = 0; 00286 } 00287 00288 #define PCL_INSTANTIATE_MaximumLikelihoodSampleConsensus(T) template class PCL_EXPORTS pcl::MaximumLikelihoodSampleConsensus<T>; 00289 00290 #endif // PCL_SAMPLE_CONSENSUS_IMPL_MLESAC_H_ 00291