voronoi_approximator.cpp

Go to the documentation of this file.

#include <algorithm>
#include <cmath>
#include <cstdlib>

#include <bwi_mapper/voronoi_approximator.h>
#include <bwi_mapper/map_utils.h>
#include <bwi_mapper/point_utils.h>

#include <opencv2/opencv_modules.hpp>
#ifdef HAVE_OPENCV_IMGPROC
// this is OpenCV 3 and we need extra includes
#include <opencv2/imgproc/imgproc.hpp>
#endif

namespace bwi_mapper {

  void VoronoiApproximator::findVoronoiPoints(double threshold, 
      bool is_naive, int sub_pixel_sampling) {

    // Get the inflated cost map
    inflateMap(threshold, map_resp_.map, inflated_map_);

    // Compute the voronoi points
    double pixel_threshold = 
      ceil(threshold / map_resp_.map.info.resolution);

    uint32_t max_dimension = 
      std::max(inflated_map_.info.height, inflated_map_.info.width);

    for (int j = 0; j < sub_pixel_sampling * (inflated_map_.info.height - 1); ++j) {
      if (j % sub_pixel_sampling == 0) {
        std::cout << "findVoronoiPoints(): On row : " << j / sub_pixel_sampling << std::endl;
      }
      for (int i = 0; i < sub_pixel_sampling * (inflated_map_.info.width - 1); 
          ++i) {

        Point2f center_pt((i + ((float)sub_pixel_sampling / 2)) / ((float)sub_pixel_sampling) + 0.001,
            (j + ((float)sub_pixel_sampling / 2)) / ((float)sub_pixel_sampling) + 0.001);

        // if (center_pt.x < inflated_map_.info.width / 2 ||
        //     i >= inflated_map_.info.width / 2 + 1)
        //   continue;
        // if (j < inflated_map_.info.height / 2 ||
        //     j >= inflated_map_.info.height / 2 + 1)
        //   continue;

        VoronoiPoint vp(lrint(center_pt.x), lrint(center_pt.y));

        // Check if this location is too close to a given obstacle
        uint32_t map_idx = MAP_IDX(inflated_map_.info.width, vp.x, vp.y);
        if (inflated_map_.data[map_idx] != 0) {
          continue;
        }

        /* std::cout << "Checking " << center_pt << ", will mark " << vp << std::endl; */

        // Use the boxes to find obstacles
        for (int box = pixel_threshold; box < max_dimension; ++box) {

          // Early termination crit - no more hope of finding more obstacles
          // at the same distance as the one found already
          if (vp.basis_points.size() != 0 &&
              box > vp.basis_distance + 1) {
            break;
          }

          // Get obstacles at this box size
          std::vector<Point2d> obstacles;
          int low_j = lrint(center_pt.y - box - 1);
          int high_j = lrint(center_pt.y + box);
          int low_i = lrint(center_pt.x - box - 1);
          int high_i = lrint(center_pt.x + box);

          /* std::cout << " - box " << box << " [" << low_i << "," << low_j << "," << high_i << "," << high_j << "]" << std::endl;  */

          // If we hit the side of the images + 1, we've seen all possible
          // sites
          if (low_i < 0 || low_j < 0 || 
              high_i >= inflated_map_.info.width 
              || high_j >= inflated_map_.info.height) {
            break;
          }

          // Corners of the box + vertical edges
          for (int j_box = low_j; j_box <= high_j; ++j_box) {
            for (int i_box = low_i; i_box <= high_i; 
                i_box += high_i - low_i) {
              uint32_t map_idx_box = 
                MAP_IDX(inflated_map_.info.width, i_box, j_box);
              bool occupied = map_resp_.map.data[map_idx_box];
              if (occupied) {
                Point2d p(i_box, j_box);
                Point2f f(i_box + 0.5, j_box + 0.5);
                p.distance_from_ref = bwi_mapper::getMagnitude(f - center_pt);
                obstacles.push_back(p);
              }
            }
          }

          // horizontal edges
          for (int j_box = low_j; j_box <= high_j;
              j_box += high_j - low_j) {
            for (int i_box = low_i + 1; i_box < high_i + 1; ++i_box) {
              uint32_t map_idx_box = 
                MAP_IDX(inflated_map_.info.width, i_box, j_box);
              bool occupied = map_resp_.map.data[map_idx_box];
              if (occupied) {
                Point2d p(i_box, j_box);
                Point2f f(i_box + 0.5, j_box + 0.5);
                p.distance_from_ref = bwi_mapper::getMagnitude(f - center_pt);
                obstacles.push_back(p);
              }
            }
          }

          // Check if any obstacles are found
          if (obstacles.size() == 0) {
            continue;
          }

          // Now that obstacles are available, sort and add to the vp as
          // necessary
          if (!is_naive) {
            std::sort(obstacles.begin(), obstacles.end(), 
                Point2dDistanceComp());
          }
          for (size_t q = 0; q < obstacles.size(); q++) {
            vp.addBasisCandidate(obstacles[q], pixel_threshold, 
                inflated_map_, is_naive);
          }
        }

        if (vp.basis_points.size() >= 2) {
          voronoi_points_.push_back(vp);
        }
        
      }
    }

    // Compute average basis distance for each voronoi point
    for (size_t i = 0; i < voronoi_points_.size(); ++i) {
      VoronoiPoint &vp = voronoi_points_[i];
      float basis_distance_sum = 0;
      for (size_t j = 0; j < vp.basis_points.size(); ++j) {
        basis_distance_sum += vp.basis_points[j].distance_from_ref;
      }
      vp.average_clearance = basis_distance_sum / vp.basis_points.size();
      std::cout << "Found VP at " << vp << " with clearance " << vp.average_clearance << std::endl;
    }

    // Label the voronoi diagram as being available
    initialized_ = true;
  }

  void VoronoiApproximator::drawOutput(cv::Mat &image) {
    if (!initialized_) {
      throw std::runtime_error("drawOutput(): voronoi diagram not "
          "initialized, call findVoronoiPoints first");
    }
    drawMap(image);
    drawMap(image, map_resp_.map.info.width);
    drawVoronoiPoints(image, map_resp_.map.info.width);
  }

  void VoronoiApproximator::printVoronoiPoints() {
    for (size_t i = 0; i < voronoi_points_.size(); ++i) {
      VoronoiPoint &vp = voronoi_points_[i];
      std::cout << " (" << vp.x << "," << vp.y << "): ";
      for (size_t j = 0; j < vp.basis_points.size(); ++j) {
        std::cout << " (" << vp.basis_points[j].x << "," 
          << vp.basis_points[j].y << "," 
          << vp.basis_points[j].distance_from_ref << "), ";
      }
      std::cout << std::endl;
    }
  }

  void VoronoiApproximator::drawVoronoiPoints(cv::Mat &image, 
      uint32_t orig_x, uint32_t orig_y) {
    for (size_t i = 0; i < voronoi_points_.size(); ++i) {
      VoronoiPoint &vp = voronoi_points_[i];
      cv::Point p(vp.x + orig_x, orig_y + vp.y);
      cv::circle(image, p, 1, cv::Scalar(255,0,0), -1, CV_AA);
      // image.at<cv::Vec3b>
      //   (orig_y + vp.y, vp.x + orig_x) = 
      //   cv::Vec3b(255,0,0);
    }
  }

} /* bwi_mapper */