correspondence_rejection_poly.hpp

/*
 * Software License Agreement (BSD License)
 *
 *  Point Cloud Library (PCL) - www.pointclouds.org
 *  Copyright (c) 2010-2011, Willow Garage, Inc.
 *  Copyright (c) 2012-, Open Perception, Inc.
 *
 *  All rights reserved.
 *
 *  Redistribution and use in source and binary forms, with or without
 *  modification, are permitted provided that the following conditions
 *  are met:
 *
 *   * Redistributions of source code must retain the above copyright
 *     notice, this list of conditions and the following disclaimer.
 *   * Redistributions in binary form must reproduce the above
 *     copyright notice, this list of conditions and the following
 *     disclaimer in the documentation and/or other materials provided
 *     with the distribution.
 *   * Neither the name of the copyright holder(s) nor the names of its
 *     contributors may be used to endorse or promote products derived
 *     from this software without specific prior written permission.
 *
 *  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 *  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 *  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
 *  FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
 *  COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
 *  INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
 *  BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 *  LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 *  CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 *  LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
 *  ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 *  POSSIBILITY OF SUCH DAMAGE.
 *
 */
 
#ifndef PCL_REGISTRATION_IMPL_CORRESPONDENCE_REJECTION_POLY_HPP_
#define PCL_REGISTRATION_IMPL_CORRESPONDENCE_REJECTION_POLY_HPP_
 
namespace pcl {
 
namespace registration {
 
template <typename SourceT, typename TargetT>
void
CorrespondenceRejectorPoly<SourceT, TargetT>::getRemainingCorrespondences(
    const pcl::Correspondences& original_correspondences,
    pcl::Correspondences& remaining_correspondences)
{
  // This is reset after all the checks below
  remaining_correspondences = original_correspondences;
 
  // Check source/target
  if (!input_) {
    PCL_ERROR("[pcl::registration::%s::getRemainingCorrespondences] No source was "
              "input! Returning all input correspondences.\n",
              getClassName().c_str());
    return;
  }
 
  if (!target_) {
    PCL_ERROR("[pcl::registration::%s::getRemainingCorrespondences] No target was "
              "input! Returning all input correspondences.\n",
              getClassName().c_str());
    return;
  }
 
  // Check cardinality
  if (cardinality_ < 2) {
    PCL_ERROR("[pcl::registration::%s::getRemainingCorrespondences] Polygon "
              "cardinality too low!. Returning all input correspondences.\n",
              getClassName().c_str());
    return;
  }
 
  // Number of input correspondences
  const int nr_correspondences = static_cast<int>(original_correspondences.size());
 
  // Not enough correspondences for polygonal rejections
  if (cardinality_ >= nr_correspondences) {
    PCL_ERROR("[pcl::registration::%s::getRemainingCorrespondences] Number of "
              "correspondences smaller than polygon cardinality! Returning all input "
              "correspondences.\n",
              getClassName().c_str());
    return;
  }
 
  // Check similarity
  if (similarity_threshold_ < 0.0f || similarity_threshold_ > 1.0f) {
    PCL_ERROR(
        "[pcl::registration::%s::getRemainingCorrespondences] Invalid edge length "
        "similarity - must be in [0,1]!. Returning all input correspondences.\n",
        getClassName().c_str());
    return;
  }
 
  // Similarity, squared
  similarity_threshold_squared_ = similarity_threshold_ * similarity_threshold_;
 
  // Initialization of result
  remaining_correspondences.clear();
  remaining_correspondences.reserve(nr_correspondences);
 
  // Number of times a correspondence is sampled and number of times it was accepted
  std::vector<int> num_samples(nr_correspondences, 0);
  std::vector<int> num_accepted(nr_correspondences, 0);
 
  // Main loop
  for (int i = 0; i < iterations_; ++i) {
    // Sample cardinality_ correspondences without replacement
    const std::vector<int> idx =
        getUniqueRandomIndices(nr_correspondences, cardinality_);
 
    // Verify the polygon similarity
    if (thresholdPolygon(original_correspondences, idx)) {
      // Increment sample counter and accept counter
      for (int j = 0; j < cardinality_; ++j) {
        ++num_samples[idx[j]];
        ++num_accepted[idx[j]];
      }
    }
    else {
      // Not accepted, only increment sample counter
      for (int j = 0; j < cardinality_; ++j)
        ++num_samples[idx[j]];
    }
  }
 
  // Now calculate the acceptance rate of each correspondence
  std::vector<float> accept_rate(nr_correspondences, 0.0f);
  for (int i = 0; i < nr_correspondences; ++i) {
    const int numsi = num_samples[i];
    if (numsi == 0)
      accept_rate[i] = 0.0f;
    else
      accept_rate[i] = static_cast<float>(num_accepted[i]) / static_cast<float>(numsi);
  }
 
  // Compute a histogram in range [0,1] for acceptance rates
  const int hist_size = nr_correspondences / 2; // TODO: Optimize this
  const std::vector<int> histogram =
      computeHistogram(accept_rate, 0.0f, 1.0f, hist_size);
 
  // Find the cut point between outliers and inliers using Otsu's thresholding method
  const int cut_idx = findThresholdOtsu(histogram);
  const float cut = static_cast<float>(cut_idx) / static_cast<float>(hist_size);
 
  // Threshold
  for (int i = 0; i < nr_correspondences; ++i)
    if (accept_rate[i] > cut)
      remaining_correspondences.push_back(original_correspondences[i]);
}
 
template <typename SourceT, typename TargetT>
std::vector<int>
CorrespondenceRejectorPoly<SourceT, TargetT>::computeHistogram(
    const std::vector<float>& data, float lower, float upper, int bins)
{
  // Result
  std::vector<int> result(bins, 0);
 
  // Last index into result and increment factor from data value --> index
  const int last_idx = bins - 1;
  const float idx_per_val = static_cast<float>(bins) / (upper - lower);
 
  // Accumulate
  for (const float& value : data)
    ++result[std::min(last_idx, static_cast<int>(value * idx_per_val))];
 
  return (result);
}
 
template <typename SourceT, typename TargetT>
int
CorrespondenceRejectorPoly<SourceT, TargetT>::findThresholdOtsu(
    const std::vector<int>& histogram)
{
  // Precision
  constexpr double eps = std::numeric_limits<double>::epsilon();
 
  // Histogram dimension
  const int nbins = static_cast<int>(histogram.size());
 
  // Mean and inverse of the number of data points
  double mean = 0.0;
  double sum_inv = 0.0;
  for (int i = 0; i < nbins; ++i) {
    mean += static_cast<double>(i * histogram[i]);
    sum_inv += static_cast<double>(histogram[i]);
  }
  sum_inv = 1.0 / sum_inv;
  mean *= sum_inv;
 
  // Probability and mean of class 1 (data to the left of threshold)
  double class_mean1 = 0.0;
  double class_prob1 = 0.0;
  double class_prob2 = 1.0;
 
  // Maximized between class variance and associated bin value
  double between_class_variance_max = 0.0;
  int result = 0;
 
  // Loop over all bin values
  for (int i = 0; i < nbins; ++i) {
    class_mean1 *= class_prob1;
 
    // Probability of bin i
    const double prob_i = static_cast<double>(histogram[i]) * sum_inv;
 
    // Class probability 1: sum of probabilities from 0 to i
    class_prob1 += prob_i;
 
    // Class probability 2: sum of probabilities from i+1 to nbins-1
    class_prob2 -= prob_i;
 
    // Avoid division by zero below
    if (std::min(class_prob1, class_prob2) < eps ||
        std::max(class_prob1, class_prob2) > 1.0 - eps)
      continue;
 
    // Class mean 1: sum of probabilities from 0 to i, weighted by bin value
    class_mean1 = (class_mean1 + static_cast<double>(i) * prob_i) / class_prob1;
 
    // Class mean 2: sum of probabilities from i+1 to nbins-1, weighted by bin value
    const double class_mean2 = (mean - class_prob1 * class_mean1) / class_prob2;
 
    // Between class variance
    const double between_class_variance = class_prob1 * class_prob2 *
                                          (class_mean1 - class_mean2) *
                                          (class_mean1 - class_mean2);
 
    // If between class variance is maximized, update result
    if (between_class_variance > between_class_variance_max) {
      between_class_variance_max = between_class_variance;
      result = i;
    }
  }
 
  return (result);
}
 
} // namespace registration
} // namespace pcl
 
#endif // PCL_REGISTRATION_IMPL_CORRESPONDENCE_REJECTION_POLY_HPP_