harris_6d.hpp

/*
 * Software License Agreement (BSD License)
 *
 *  Point Cloud Library (PCL) - www.pointclouds.org
 *  Copyright (c) 2010-2011, Willow Garage, 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 Willow Garage, Inc. 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_HARRIS_KEYPOINT_6D_IMPL_H_
#define PCL_HARRIS_KEYPOINT_6D_IMPL_H_
 
#include <Eigen/Eigenvalues> // for SelfAdjointEigenSolver
#include <pcl/keypoints/harris_6d.h>
#include <pcl/common/io.h>
#include <pcl/features/normal_3d.h>
//#include <pcl/features/fast_intensity_gradient.h>
#include <pcl/features/intensity_gradient.h>
#include <pcl/features/integral_image_normal.h>
 
template <typename PointInT, typename PointOutT, typename NormalT> void
pcl::HarrisKeypoint6D<PointInT, PointOutT, NormalT>::setThreshold (float threshold)
{
  threshold_= threshold;
}
 
template <typename PointInT, typename PointOutT, typename NormalT> void
pcl::HarrisKeypoint6D<PointInT, PointOutT, NormalT>::setRadius (float radius)
{
  search_radius_ = radius;
}
 
template <typename PointInT, typename PointOutT, typename NormalT> void
pcl::HarrisKeypoint6D<PointInT, PointOutT, NormalT>::setRefine (bool do_refine)
{
  refine_ = do_refine;
}
 
template <typename PointInT, typename PointOutT, typename NormalT> void
pcl::HarrisKeypoint6D<PointInT, PointOutT, NormalT>::setNonMaxSupression (bool nonmax)
{
  nonmax_ = nonmax;
}
 
//////////////////////////////////////////////////////////////////////////////////////////////////////////////////
template <typename PointInT, typename PointOutT, typename NormalT> void
pcl::HarrisKeypoint6D<PointInT, PointOutT, NormalT>::calculateCombinedCovar (const pcl::Indices& neighbors, float* coefficients) const
{
  std::fill_n(coefficients, 21, 0);
  unsigned count = 0;
  for (const auto &neighbor : neighbors)
  {
    if (std::isfinite ((*normals_)[neighbor].normal_x) && std::isfinite ((*intensity_gradients_)[neighbor].gradient [0]))
    {
      coefficients[ 0] += (*normals_)[neighbor].normal_x * (*normals_)[neighbor].normal_x;
      coefficients[ 1] += (*normals_)[neighbor].normal_x * (*normals_)[neighbor].normal_y;
      coefficients[ 2] += (*normals_)[neighbor].normal_x * (*normals_)[neighbor].normal_z;
      coefficients[ 3] += (*normals_)[neighbor].normal_x * (*intensity_gradients_)[neighbor].gradient [0];
      coefficients[ 4] += (*normals_)[neighbor].normal_x * (*intensity_gradients_)[neighbor].gradient [1];
      coefficients[ 5] += (*normals_)[neighbor].normal_x * (*intensity_gradients_)[neighbor].gradient [2];
 
      coefficients[ 6] += (*normals_)[neighbor].normal_y * (*normals_)[neighbor].normal_y;
      coefficients[ 7] += (*normals_)[neighbor].normal_y * (*normals_)[neighbor].normal_z;
      coefficients[ 8] += (*normals_)[neighbor].normal_y * (*intensity_gradients_)[neighbor].gradient [0];
      coefficients[ 9] += (*normals_)[neighbor].normal_y * (*intensity_gradients_)[neighbor].gradient [1];
      coefficients[10] += (*normals_)[neighbor].normal_y * (*intensity_gradients_)[neighbor].gradient [2];
 
      coefficients[11] += (*normals_)[neighbor].normal_z * (*normals_)[neighbor].normal_z;
      coefficients[12] += (*normals_)[neighbor].normal_z * (*intensity_gradients_)[neighbor].gradient [0];
      coefficients[13] += (*normals_)[neighbor].normal_z * (*intensity_gradients_)[neighbor].gradient [1];
      coefficients[14] += (*normals_)[neighbor].normal_z * (*intensity_gradients_)[neighbor].gradient [2];
 
      coefficients[15] += (*intensity_gradients_)[neighbor].gradient [0] * (*intensity_gradients_)[neighbor].gradient [0];
      coefficients[16] += (*intensity_gradients_)[neighbor].gradient [0] * (*intensity_gradients_)[neighbor].gradient [1];
      coefficients[17] += (*intensity_gradients_)[neighbor].gradient [0] * (*intensity_gradients_)[neighbor].gradient [2];
 
      coefficients[18] += (*intensity_gradients_)[neighbor].gradient [1] * (*intensity_gradients_)[neighbor].gradient [1];
      coefficients[19] += (*intensity_gradients_)[neighbor].gradient [1] * (*intensity_gradients_)[neighbor].gradient [2];
 
      coefficients[20] += (*intensity_gradients_)[neighbor].gradient [2] * (*intensity_gradients_)[neighbor].gradient [2];
 
      ++count;
    }
  }
  if (count > 0)
  {
    float norm = 1.0 / static_cast<float>(count);
    coefficients[ 0] *= norm;
    coefficients[ 1] *= norm;
    coefficients[ 2] *= norm;
    coefficients[ 3] *= norm;
    coefficients[ 4] *= norm;
    coefficients[ 5] *= norm;
    coefficients[ 6] *= norm;
    coefficients[ 7] *= norm;
    coefficients[ 8] *= norm;
    coefficients[ 9] *= norm;
    coefficients[10] *= norm;
    coefficients[11] *= norm;
    coefficients[12] *= norm;
    coefficients[13] *= norm;
    coefficients[14] *= norm;
    coefficients[15] *= norm;
    coefficients[16] *= norm;
    coefficients[17] *= norm;
    coefficients[18] *= norm;
    coefficients[19] *= norm;
    coefficients[20] *= norm;
  }
}
 
//////////////////////////////////////////////////////////////////////////////////////////////////////////////////
template <typename PointInT, typename PointOutT, typename NormalT> void
pcl::HarrisKeypoint6D<PointInT, PointOutT, NormalT>::detectKeypoints (PointCloudOut &output)
{
  if (normals_->empty ())
  {
    normals_->reserve (surface_->size ());
    if (!surface_->isOrganized ())
    {
      pcl::NormalEstimation<PointInT, NormalT> normal_estimation;
      normal_estimation.setInputCloud (surface_);
      normal_estimation.setRadiusSearch (search_radius_);
      normal_estimation.compute (*normals_);
    }
    else
    {
      IntegralImageNormalEstimation<PointInT, NormalT> normal_estimation;
      normal_estimation.setNormalEstimationMethod (pcl::IntegralImageNormalEstimation<PointInT, NormalT>::SIMPLE_3D_GRADIENT);
      normal_estimation.setInputCloud (surface_);
      normal_estimation.setNormalSmoothingSize (5.0);
      normal_estimation.compute (*normals_);
    }
  }
 
  pcl::PointCloud<pcl::PointXYZI>::Ptr cloud (new pcl::PointCloud<pcl::PointXYZI>);
  cloud->resize (surface_->size ());
#pragma omp parallel for \
  default(none) \
  num_threads(threads_)
  for (unsigned idx = 0; idx < surface_->size (); ++idx)
  {
    cloud->points [idx].x = surface_->points [idx].x;
    cloud->points [idx].y = surface_->points [idx].y;
    cloud->points [idx].z = surface_->points [idx].z;
    //grayscale = 0.2989 * R + 0.5870 * G + 0.1140 * B
 
    cloud->points [idx].intensity = 0.00390625 * (0.114 * float(surface_->points [idx].b) + 0.5870 * float(surface_->points [idx].g) + 0.2989 * float(surface_->points [idx].r));
  }
  pcl::copyPointCloud (*surface_, *cloud);
 
  IntensityGradientEstimation<PointXYZI, NormalT, IntensityGradient> grad_est;
  grad_est.setInputCloud (cloud);
  grad_est.setInputNormals (normals_);
  grad_est.setRadiusSearch (search_radius_);
  grad_est.compute (*intensity_gradients_);
  
#pragma omp parallel for \
  default(none) \
  num_threads(threads_)
  for (std::size_t idx = 0; idx < intensity_gradients_->size (); ++idx)
  {
    float len = intensity_gradients_->points [idx].gradient_x * intensity_gradients_->points [idx].gradient_x +
                intensity_gradients_->points [idx].gradient_y * intensity_gradients_->points [idx].gradient_y +
                intensity_gradients_->points [idx].gradient_z * intensity_gradients_->points [idx].gradient_z ;
 
    // Suat: ToDo: remove this magic number or expose using set/get
    if (len > 200.0)
    {
      len = 1.0 / std::sqrt (len);
      intensity_gradients_->points [idx].gradient_x *= len;
      intensity_gradients_->points [idx].gradient_y *= len;
      intensity_gradients_->points [idx].gradient_z *= len;
    }
    else
    {
      intensity_gradients_->points [idx].gradient_x = 0;
      intensity_gradients_->points [idx].gradient_y = 0;
      intensity_gradients_->points [idx].gradient_z = 0;
    }
  }
 
  typename pcl::PointCloud<PointOutT>::Ptr response (new pcl::PointCloud<PointOutT>);
  response->points.reserve (input_->size());
  responseTomasi(*response);
 
  // just return the response
  if (!nonmax_)
  {
    output = *response;
    // we do not change the denseness in this case
    output.is_dense = input_->is_dense;
    for (std::size_t i = 0; i < response->size (); ++i)
      keypoints_indices_->indices.push_back (i);
  }
  else
  {
    output.clear ();
    output.reserve (response->size());
 
#pragma omp parallel for \
  default(none) \
  num_threads(threads_)
    for (std::size_t idx = 0; idx < response->size (); ++idx)
    {
      if (!isFinite ((*response)[idx]) || (*response)[idx].intensity < threshold_)
        continue;
 
      pcl::Indices nn_indices;
      std::vector<float> nn_dists;
      tree_->radiusSearch (idx, search_radius_, nn_indices, nn_dists);
      bool is_maxima = true;
      for (const auto& index : nn_indices)
      {
        if ((*response)[idx].intensity < (*response)[index].intensity)
        {
          is_maxima = false;
          break;
        }
      }
      if (is_maxima)
        #pragma omp critical
      {
        output.push_back ((*response)[idx]);
        keypoints_indices_->indices.push_back (idx);
      }
    }
 
    if (refine_)
      refineCorners (output);
 
    output.height = 1;
    output.width = output.size();
    output.is_dense = true;
  }
}
 
template <typename PointInT, typename PointOutT, typename NormalT> void
pcl::HarrisKeypoint6D<PointInT, PointOutT, NormalT>::responseTomasi (PointCloudOut &output) const
{
  // get the 6x6 covar-mat
  PointOutT pointOut;
  PCL_ALIGN (16) float covar [21];
  Eigen::SelfAdjointEigenSolver <Eigen::Matrix<float, 6, 6> > solver;
  Eigen::Matrix<float, 6, 6> covariance;
 
#pragma omp parallel for \
  default(none) \
  firstprivate(pointOut, covar, covariance, solver) \
  num_threads(threads_)
  for (unsigned pIdx = 0; pIdx < input_->size (); ++pIdx)
  {
    const PointInT& pointIn = input_->points [pIdx];
    pointOut.intensity = 0.0; //std::numeric_limits<float>::quiet_NaN ();
    if (isFinite (pointIn))
    {
      pcl::Indices nn_indices;
      std::vector<float> nn_dists;
      tree_->radiusSearch (pointIn, search_radius_, nn_indices, nn_dists);
      calculateCombinedCovar (nn_indices, covar);
 
      float trace = covar [0] + covar [6] + covar [11] + covar [15] + covar [18] + covar [20];
      if (trace != 0)
      {
        covariance.coeffRef ( 0) = covar [ 0];
        covariance.coeffRef ( 1) = covar [ 1];
        covariance.coeffRef ( 2) = covar [ 2];
        covariance.coeffRef ( 3) = covar [ 3];
        covariance.coeffRef ( 4) = covar [ 4];
        covariance.coeffRef ( 5) = covar [ 5];
 
        covariance.coeffRef ( 7) = covar [ 6];
        covariance.coeffRef ( 8) = covar [ 7];
        covariance.coeffRef ( 9) = covar [ 8];
        covariance.coeffRef (10) = covar [ 9];
        covariance.coeffRef (11) = covar [10];
 
        covariance.coeffRef (14) = covar [11];
        covariance.coeffRef (15) = covar [12];
        covariance.coeffRef (16) = covar [13];
        covariance.coeffRef (17) = covar [14];
 
        covariance.coeffRef (21) = covar [15];
        covariance.coeffRef (22) = covar [16];
        covariance.coeffRef (23) = covar [17];
 
        covariance.coeffRef (28) = covar [18];
        covariance.coeffRef (29) = covar [19];
 
        covariance.coeffRef (35) = covar [20];
 
        covariance.coeffRef ( 6) = covar [ 1];
 
        covariance.coeffRef (12) = covar [ 2];
        covariance.coeffRef (13) = covar [ 7];
 
        covariance.coeffRef (18) = covar [ 3];
        covariance.coeffRef (19) = covar [ 8];
        covariance.coeffRef (20) = covar [12];
 
        covariance.coeffRef (24) = covar [ 4];
        covariance.coeffRef (25) = covar [ 9];
        covariance.coeffRef (26) = covar [13];
        covariance.coeffRef (27) = covar [16];
 
        covariance.coeffRef (30) = covar [ 5];
        covariance.coeffRef (31) = covar [10];
        covariance.coeffRef (32) = covar [14];
        covariance.coeffRef (33) = covar [17];
        covariance.coeffRef (34) = covar [19];
 
        solver.compute (covariance);
        pointOut.intensity = solver.eigenvalues () [3];
      }
    }
 
    pointOut.x = pointIn.x;
    pointOut.y = pointIn.y;
    pointOut.z = pointIn.z;
 
    #pragma omp critical
    output.push_back(pointOut);
  }
  output.height = input_->height;
  output.width = input_->width;
}
 
template <typename PointInT, typename PointOutT, typename NormalT> void
pcl::HarrisKeypoint6D<PointInT, PointOutT, NormalT>::refineCorners (PointCloudOut &corners) const
{
  pcl::search::KdTree<PointInT> search;
  search.setInputCloud(surface_);
 
  Eigen::Matrix3f nnT;
  Eigen::Matrix3f NNT;
  Eigen::Vector3f NNTp;
  const Eigen::Vector3f* normal;
  const Eigen::Vector3f* point;
  float diff;
  const unsigned max_iterations = 10;
  for (typename PointCloudOut::iterator cornerIt = corners.begin(); cornerIt != corners.end(); ++cornerIt)
  {
    unsigned iterations = 0;
    do {
      NNT.setZero();
      NNTp.setZero();
      PointInT corner;
      corner.x = cornerIt->x;
      corner.y = cornerIt->y;
      corner.z = cornerIt->z;
      pcl::Indices nn_indices;
      std::vector<float> nn_dists;      
      search.radiusSearch (corner, search_radius_, nn_indices, nn_dists);
      for (const auto& index : nn_indices)
      {
        normal = reinterpret_cast<const Eigen::Vector3f*> (&((*normals_)[index].normal_x));
        point = reinterpret_cast<const Eigen::Vector3f*> (&((*surface_)[index].x));
        nnT = (*normal) * (normal->transpose());
        NNT += nnT;
        NNTp += nnT * (*point);
      }
      if (NNT.determinant() != 0)
        *(reinterpret_cast<Eigen::Vector3f*>(&(cornerIt->x))) = NNT.inverse () * NNTp;
 
      diff = (cornerIt->x - corner.x) * (cornerIt->x - corner.x) +
             (cornerIt->y - corner.y) * (cornerIt->y - corner.y) +
             (cornerIt->z - corner.z) * (cornerIt->z - corner.z);
 
    } while (diff > 1e-6 && ++iterations < max_iterations);
  }
}
 
#define PCL_INSTANTIATE_HarrisKeypoint6D(T,U,N) template class PCL_EXPORTS pcl::HarrisKeypoint6D<T,U,N>;
#endif // #ifndef PCL_HARRIS_KEYPOINT_6D_IMPL_H_