Convex Hull filtering by simple bounding box

[pseandersson]

A pointcloud consist of 8 points, making up a bounding box in 3D. I wanted to select all points inside the boundingbox in another point cloud by computing a convex hull of the points and filtering the cloud by the crop hull filter in 3 dimensions. The result changed when I duplicated two points in the cloud. In both cases, approximatley one half of the points were selected.

[jspricke]

Hi,

could you please provide a minimal test case, so we can reproduce thid?

Thanks

Jochen

[pseandersson]

// As a main cloud "cloud" 0000002.txt is used from KITTI Benchmark 2011_09_26_drive_0001
// available at http://www.cvlibs.net/datasets/kitti/raw_data.php
/* getBoundingBox(...)->points()
22.9035 7.80552 -1.79216
27.4452 7.61189 -1.79216
27.5215 9.40126 -1.79216
22.9798 9.59489 -1.79216
22.9035 7.80552 -1.79216 // Duplicate of row 1
22.9035 7.80552 0.483439
27.4452 7.61189 0.483439
27.5215 9.40126 0.483439
22.9798 9.59489 0.483439
22.9035 7.80552 0.483439 // Duplicate of row 6
*/
pcl::PointCloudpcl::PointXYZ::Ptr boundingbox_ptr, surface_hull (new pcl::PointCloudpcl::PointXYZ);
boundingbox_ptr = getBoundingBox(...);

pcl::ConvexHullpcl::PointXYZ hull;
hull.setInputCloud(boundingbox_ptr);
hull.setDimension(3);
std::vectorpcl::Vertices polygons;
hull.reconstruct(*surface_hull,polygons);

pcl::PointCloudpcl::PointXYZ::Ptr objects (new pcl::PointCloudpcl::PointXYZ);
pcl::CropHullpcl::PointXYZ bb_filter;

bb_filter.setDim(3);
bb_filter.setInputCloud(cloud);
bb_filter.setHullIndices(polygons);
bb_filter.setHullCloud(boundingbox_ptr);
bb_filter.filter(*objects);

[jspricke]

Here is a minimal example:

#include <iostream>

#include <pcl/point_cloud.h>
#include <pcl/point_types.h>
#include <pcl/surface/convex_hull.h>
#include <pcl/filters/crop_hull.h>

int main()
{
  pcl::PointCloud<pcl::PointXYZ>::Ptr cloud (new pcl::PointCloud<pcl::PointXYZ>);
  cloud->push_back(pcl::PointXYZ(25.587, 7.8600001, 0.29100001));
  cloud->push_back(pcl::PointXYZ(24.372999, 8.5909996, 0.28799999));

  pcl::PointCloud<pcl::PointXYZ>::Ptr boundingbox_ptr (new pcl::PointCloud<pcl::PointXYZ>);
  boundingbox_ptr->push_back(pcl::PointXYZ(22.9035, 7.80552, -1.79216));
  boundingbox_ptr->push_back(pcl::PointXYZ(27.4452, 7.61189, -1.79216));
  boundingbox_ptr->push_back(pcl::PointXYZ(27.5215, 9.40126, -1.79216));
  boundingbox_ptr->push_back(pcl::PointXYZ(22.9798, 9.59489, -1.79216));
  boundingbox_ptr->push_back(pcl::PointXYZ(22.9035, 7.80552, 0.483439));
  boundingbox_ptr->push_back(pcl::PointXYZ(27.4452, 7.61189, 0.483439));
  boundingbox_ptr->push_back(pcl::PointXYZ(27.5215, 9.40126, 0.483439));
  boundingbox_ptr->push_back(pcl::PointXYZ(22.9798, 9.59489, 0.483439));

  pcl::ConvexHull<pcl::PointXYZ> hull;
  hull.setInputCloud(boundingbox_ptr);
  hull.setDimension(3);
  std::vector<pcl::Vertices> polygons;

  pcl::PointCloud<pcl::PointXYZ>::Ptr surface_hull (new pcl::PointCloud<pcl::PointXYZ>);
  hull.reconstruct(*surface_hull, polygons);

  for(int i = 0; i < polygons.size(); i++)
    std::cout << polygons[i] << std::endl;

  pcl::PointCloud<pcl::PointXYZ>::Ptr objects (new pcl::PointCloud<pcl::PointXYZ>);
  pcl::CropHull<pcl::PointXYZ> bb_filter;

  bb_filter.setDim(3);
  bb_filter.setInputCloud(cloud);
  bb_filter.setHullIndices(polygons);
  bb_filter.setHullCloud(boundingbox_ptr);
  bb_filter.filter(*objects);
  std::cout << objects->size() << std::endl;

  boundingbox_ptr->clear();
  objects->clear();

  boundingbox_ptr->push_back(pcl::PointXYZ(22.9035, 7.80552, -1.79216));
  boundingbox_ptr->push_back(pcl::PointXYZ(27.4452, 7.61189, -1.79216));
  boundingbox_ptr->push_back(pcl::PointXYZ(27.5215, 9.40126, -1.79216));
  boundingbox_ptr->push_back(pcl::PointXYZ(22.9798, 9.59489, -1.79216));
  boundingbox_ptr->push_back(pcl::PointXYZ(22.9035, 7.80552, -1.79216)); // Duplicate of row 1
  boundingbox_ptr->push_back(pcl::PointXYZ(22.9035, 7.80552, 0.483439));
  boundingbox_ptr->push_back(pcl::PointXYZ(27.4452, 7.61189, 0.483439));
  boundingbox_ptr->push_back(pcl::PointXYZ(27.5215, 9.40126, 0.483439));
  boundingbox_ptr->push_back(pcl::PointXYZ(22.9798, 9.59489, 0.483439));

  bb_filter.setHullCloud(boundingbox_ptr);
  bb_filter.filter(*objects);
  std::cout << objects->size() << std::endl;
}

Problem seems to be that the indices in polygons don't correspond to the ones in boundingbox_ptr.

[rbrusu]

Is there any follow up needed on this issue?

[rbrusu]

Closed due to lack of activity.

[gtinchev]

The example by @jspricke is wrong.
It should be:
bb_filter.setHullCloud(surface_hull);

As one should be using the points after the reconstruction.