narf_keypoint_extraction.rst

How to extract NARF keypoint from a range image

This tutorial demonstrates how to extract NARF key points from a range image. The executable enables us to load a point cloud from disc (or create it if not given), extract interest points on it and visualize the result, both in an image and a 3D viewer.

The code

First, create a file called, let's say, narf_keypoint_extraction.cpp in your favorite editor, and place the following code inside it:

.. literalinclude:: sources/narf_keypoint_extraction/narf_keypoint_extraction.cpp
   :language: cpp
   :linenos:

Explanation

In the beginning we do command line parsing, read a point cloud from disc (or create it if not provided), create a range image and visualize it. All of these steps are already covered in the previous tutorial :ref:`range_image_visualization` Range image visualization.

The interesting part begins here:

...
pcl::RangeImageBorderExtractor range_image_border_extractor;
pcl::NarfKeypoint narf_keypoint_detector (&range_image_border_extractor);
narf_keypoint_detector.setRangeImage (&range_image);
narf_keypoint_detector.getParameters ().support_size = support_size;
//narf_keypoint_detector.getParameters ().add_points_on_straight_edges = true;
//narf_keypoint_detector.getParameters ().distance_for_additional_points = 0.5;

pcl::PointCloud<int> keypoint_indices;
narf_keypoint_detector.compute (keypoint_indices);
std::cout << "Found "<<keypoint_indices.points.size ()<<" key points.\n";
...

This creates a RangeImageBorderExtractor object, that is needed for the interest point extraction. If you are interested in this you can have a look at the Range Image Border Extraction tutorial. In this case we just use the RangeImageBorderExtractor object with its default parameters. Then we create the NarfKeypoint object, give it the RangeImageBorderExtractor object, the range image and set the support size (the size of the sphere around a point that includes points that are used for the determination of the interest value). The commented out part contains some parameters that you can test out if you want. Next we create the object where the indices of the determined keypoints will be saved and compute them. In the last step we output the number of found keypoints.

The remaining code just visualizes the results in a range image widget and also in a 3D viewer.

Compiling and running the program

Add the following lines to your CMakeLists.txt file:

.. literalinclude:: sources/narf_keypoint_extraction/CMakeLists.txt
   :language: cmake
   :linenos:

After you have made the executable, you can run it. Simply do:

$ ./narf_keypoint_extraction -m

This will use an autogenerated point cloud of a rectangle floating in space. The key points are detected in the corners. The parameter -m is necessary, since the area around the rectangle is unseen and therefore the system can not detect it as a border. The option -m changes the unseen area to maximum range readings, thereby enabling the system to use these borders.

You can also try it with a point cloud file from your hard drive:

$ ./narf_keypoint_extraction <point_cloud.pcd>

The output should look similar to this: