Estimating Surface Normals in a PointCloud#

In this tutorial, we will learn how to obtain the surface normals of each point in the cloud.

Note

This tutorial is applicable for execution for 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>/normal_estimation

  2. oneapi_normal_estimation.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/io/pcd_io.h>
 4#include <pcl/oneapi/features/normal_3d.h>
 5#include <pcl/oneapi/kdtree/kdtree_flann.h>
 6#include <pcl/oneapi/point_cloud.h>
 7
 8int main (int argc, char** argv)
 9{
10  int k = 10;
11  float radius = 0.01;
12
13  std::cout << "Running on device: " << dpct::get_default_queue().get_device().get_info<sycl::info::device::name>() << "\n";
14
15  // load point cloud
16  pcl::PointCloud<pcl::PointXYZ>::Ptr cloud_ptr( new pcl::PointCloud<pcl::PointXYZ>() );
17
18  int result = pcl::io::loadPCDFile("bun0.pcd", *cloud_ptr);
19  if (result != 0)
20  {
21    pcl::console::print_info ("Load pcd file failed.\n");
22    return result;
23  }
24
25  // estimate normals with knn search
26  pcl::oneapi::NormalEstimation<pcl::PointXYZ, pcl::Normal> ne;
27  ne.setSearchMethod (pcl::oneapi::KdTreeFLANN<pcl::PointXYZ>::Ptr (new pcl::oneapi::KdTreeFLANN<pcl::PointXYZ>));
28  ne.setInputCloud(cloud_ptr);
29  ne.setKSearch(k);
30
31  // save normal estimation to CPU memory point cloud
32  pcl::PointCloud<pcl::Normal>::Ptr normals_knn(new pcl::PointCloud<pcl::Normal>);
33  ne.compute(*normals_knn);
34
35  std::cout << "normals_knn.size (): " << normals_knn->size () << std::endl;
36
37  // estimate normals with radius search
38  ne.setSearchMethod (pcl::oneapi::KdTreeFLANN<pcl::PointXYZ>::Ptr (new pcl::oneapi::KdTreeFLANN<pcl::PointXYZ>));
39  ne.setInputCloud(cloud_ptr);
40  ne.setRadiusSearch(radius);
41  ne.setKSearch(0);
42
43  // save normal estimation output to device shared memory point cloud
44  pcl::oneapi::PointCloudDev<pcl::Normal>::Ptr normals_radius(new pcl::oneapi::PointCloudDev<pcl::Normal>) ;
45  ne.compute(*normals_radius);
46
47  std::cout << "normals_radius.size (): " << normals_radius->size () << std::endl;
48
49  return 0;
50}

  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_normal_estimation

  7. Expected results example:

    normals_knn.size (): 397
normals_radius.size (): 397

Code Explanation#

The example PCD is initially loaded into a PointCloud<PointXYZ>.

// load point cloud
pcl::PointCloud<pcl::PointXYZ>::Ptr cloud_ptr( new pcl::PointCloud<pcl::PointXYZ>() );

int result = pcl::io::loadPCDFile("bun0.pcd", *cloud_ptr);
if (result != 0)
{
  pcl::console::print_info ("Load pcd file failed.\n");
  return result;
}

This tutorial includes two normal estimation processes: KNN search and Radius search. The initial step involves adjusting the parameters for normal estimation in the KNN search.

// estimate normals with knn search
pcl::oneapi::NormalEstimation<pcl::PointXYZ, pcl::Normal> ne;
ne.setSearchMethod (pcl::oneapi::KdTreeFLANN<pcl::PointXYZ>::Ptr (new pcl::oneapi::KdTreeFLANN<pcl::PointXYZ>));
ne.setInputCloud(cloud_ptr);
ne.setKSearch(k);

Normal estimation is then executed.

pcl::PointCloud<pcl::Normal>::Ptr normals_knn(new pcl::PointCloud<pcl::Normal>);
ne.compute(*normals_knn);

The parameters for normal estimation are modified for the radius search.

// estimate normals with radius search
ne.setSearchMethod (pcl::oneapi::KdTreeFLANN<pcl::PointXYZ>::Ptr (new pcl::oneapi::KdTreeFLANN<pcl::PointXYZ>));
ne.setInputCloud(cloud_ptr);
ne.setRadiusSearch(radius);
ne.setKSearch(0);

Normal estimation is performed once more.

// save normal estimation output to device shared memory point cloud
pcl::oneapi::PointCloudDev<pcl::Normal>::Ptr normals_radius(new pcl::oneapi::PointCloudDev<pcl::Normal>) ;
ne.compute(*normals_radius);