How to Train a Custom Point Cloud Classification in Global Mapper Pro
Are you looking to create point cloud classifications to identify unique features in your point clouds? Then you are in the right place!
Creating custom feature classifications is a cutting-edge new tool in version 25 of Global Mapper Pro. It’s based on segmentation, a more manual classification method, which finds objects in a point cloud by looking at the attributes and structure of the points. For example, to segment paint stripes on the road, you would look for points that make up a flat surface, have the same color, etc. This method operates on the assumption that each “object” in the point cloud, each cluster of points, has a signature made up of attributes and/or structures that separate it from its neighbors.
The new Custom Point Cloud Classification tool takes advantage of these signatures and lets you train a custom classification tool to look for a specific signature in your point cloud. Once the classification has been successfully created, other point clouds can be loaded into this workspace to be classified. This tool is currently in beta, but keep an eye out for updates to come with Global Mapper’s continuous development.
This blog will walk through a basic workflow of creating a custom classification for identifying these airplanes. To try other workflows like this for yourself with more detailed step-by-step instructions, see our free online training portal. The online courses provide workflows for the new features in the most recent release of Global Mapper.
Tip* Before running your new classification, use the existing classification tools to identify and filter out other objects in the point cloud to omit them from custom classification. This will increase accuracy by narrowing down the points the custom classification needs to parse.
This step is technically optional but highly recommended. The custom classification tool looks at many of the same attributes as the segmentation tool. When training the classification, selecting points by segment will help to ensure that all points have similar signatures. For more information on segmentation, see the blogs listed below. The dialog boxes are old; they reflect earlier versions, but the settings are unchanged:
Checking this box will open the Training Samples window and add new options below the check box. Click Add Class to generate and name your new custom point cloud class. This name will appear in the dropdown and also in the Configuration menu under lidar. That is also where you can change the assigned color and edit the name in the future.
This option also enables the ability to train existing classifications: ground, building, etc. Instead of starting from scratch with a new classification, you can train the existing classes to identify specific feature types, such as buildings with a unique shape. In the Classification and Extraction Shared Settings section, expand a classification and click Train to begin the process.
The Feature Model plot is a way to visualize the signatures of your custom classes. Use this tool to investigate why some features are or aren’t being captured or compare the signature to existing classes to verify that your classification looks for unique points. This bar graph highlights a few of the variables assessed when searching for features in a point cloud. Point cloud segments with value characteristics that fall within these ranges will be classified.
The example shown here with pink lines (far right of each section) uses a custom classification “Trucks” trained to identify semi trucks in aerial lidar. This is compared to ground and building points. Using the Feature Model Plot, we can see that the signature of these features is like a narrower subset of “Buildings.” The Z normals are up near 1 because the roofs of trucks are all flat. The X and Y values are more variable, probably because this was trained on a nadir, aerial data set that doesn’t have many oblique angles. It’s older, low-resolution data that only captured a few points on the sides of the trucks.