Downsampling 3D Point Clouds with a Voxelized Grid#

This tutorial covers the process of downsampling / reducing the number of points in a 3D point cloud through a voxelized grid approach.

Note

This tutorial is applicable for execution both within inside and outside a Docker image. It assumes that the pcl-oneapi-tutorials Deb package is installed, and the user has copied the tutorial directory from /opt/intel/pcl/oneapi/tutorials/ to a user-writable directory.

  1. Prepare the environment:

    cd <path-to-oneapi-tutorials>/voxel_grid

  2. oneapi_voxel_grid.cpp should be in the directory with following content:

     1// SPDX-License-Identifier: Apache-2.0
 2// Copyright (C) 2025 Intel Corporation
 3#include <pcl/oneapi/filters/voxel_grid.h>
 4#include <pcl/io/pcd_io.h>
 5#include <pcl/point_types.h>
 6#include <pcl/point_cloud.h>
 7
 8
 9int main (int argc, char** argv)
10{
11  std::cout << "Running on device: " << dpct::get_default_queue().get_device().get_info<sycl::info::device::name>() << "\n";
12
13  // Read Point Cloud
14  pcl::PointCloud<pcl::PointXYZ>::Ptr cloud_( new pcl::PointCloud<pcl::PointXYZ>() );
15  int result = pcl::io::loadPCDFile("table_scene_lms400.pcd", *cloud_);
16  if (result != 0)
17  {
18    pcl::console::print_info ("Load pcd file failed.\n");
19    return result;
20  }
21
22  // Prepare Point Cloud Memory (output)
23  pcl::PointCloud<pcl::PointXYZ>::Ptr cloud_filtered_( new pcl::PointCloud<pcl::PointXYZ>() );
24
25  // GPU calculate
26  pcl::oneapi::VoxelGrid<pcl::PointXYZ> vg_oneapi;
27  vg_oneapi.setInputCloud(cloud_);
28  float leafsize= 0.005f;
29  vg_oneapi.setLeafSize (leafsize, leafsize, leafsize);
30  vg_oneapi.filter(*cloud_filtered_);
31
32  // print log
33  std::cout << "[oneapi voxel grid] PointCloud before filtering: " << cloud_->size() << std::endl;
34  std::cout << "[oneapi voxel grid] PointCloud after filtering: " << cloud_filtered_->size() << std::endl;
35}

  3. Source the Intel® oneAPI Base Toolkit environment:

    source /opt/intel/oneapi/setvars.sh

  4. (Optional) Set up proxy setting to download test data:

    export http_proxy="http://<http_proxy>:port"
export https_proxy="http://<https_proxy>:port"

  5. Build the code:

    mkdir build && cd build
cmake ../
make -j

  6. Run the binary:

    ./oneapi_voxel_grid

  7. Expected results example:

    [oneapi voxel grid] PointCloud before filtering: 460400
[oneapi voxel grid] PointCloud after filtering: 141525

Code Explanation#

Now, let’s explain the code in detail.

First, load the example PCD into a PointCloud<PointXYZ>.

// Read Point Cloud
pcl::PointCloud<pcl::PointXYZ>::Ptr cloud_( new pcl::PointCloud<pcl::PointXYZ>() );
int result = pcl::io::loadPCDFile("table_scene_lms400.pcd", *cloud_);
if (result != 0)
{
  pcl::console::print_info ("Load pcd file failed.\n");
  return result;
}

Then prepare output buffer for filtered result.

// Prepare Point Cloud Memory (output)
pcl::PointCloud<pcl::PointXYZ>::Ptr cloud_filtered_( new pcl::PointCloud<pcl::PointXYZ>() );

Next, starts to compute the result.

// GPU calculate
pcl::oneapi::VoxelGrid<pcl::PointXYZ> vg_oneapi;
vg_oneapi.setInputCloud(cloud_);
float leafsize= 0.005f;
vg_oneapi.setLeafSize (leafsize, leafsize, leafsize);
vg_oneapi.filter(*cloud_filtered_);

Result(output log).

// print log
std::cout << "[oneapi voxel grid] PointCloud before filtering: " << cloud_->size() << std::endl;
std::cout << "[oneapi voxel grid] PointCloud after filtering: " << cloud_filtered_->size() << std::endl;