Automated Point Cloud Analysis – Global Mapper

Find the Automatic Point Cloud Analysis tool in the Classification dropdown from the Lidar toolbar.

This function requires Global Mapper Professional

The Automatic Point Cloud Analysis Tool is the hub of point cloud classification, segmentation, and extraction tools in Global Mapper Pro. From this tool, you can choose which layers are to be operated on, filter or limit processing, and handle feature models for classification. As with other editing tools in Global Mapper, the original point cloud files will not be edited, but changes to the files will be saved in the workspace or preserved in export of the Lidar data once classified.

In this tool, functions and settings are built into separate sections:

Input Configuration

Input Layer - Choose which loaded layers will be operated on.
- Selected Only- Only the points currently selected in the workspace by the digitizer tool will be operated on.
Specify Bounds... - Set the bounds for the classification by drawing a box on the workspace, by using coordinate extents, or by cropping to a selected area feature. When classifying a large dataset, it is recommended to test the classification on a sample area. Run the classifications multiple times on the sample area, adjusting to values, to obtain a good classification result prior to processing the entire data extent.
Filter Points... - Contains additional settings for filtering points for classification by elevation and color values, Source ID, and existing classification. The user may also exclude all points outside of a specified scan angle.

Input Statistics

Displays Point Cloud statistics for the enabled points/point clouds. This provides a numeric snapshot of the data being processed.
Additional Point Cloud Statistics can be found in the layer’s metadata.

Live Classification Preview

Check the box to enable a 3D viewer window to display a preview of the results of your current classification settings.

Point Selection: Use the Digitizer to select points in the main map view. You may select a new group of points at any time, and the preview will update live.
Live Update: As you enable / disable classification models and adjust the settings, the preview will update. If a large number of points are selected, it may take time for the preview to update. A progress bar will display in the Classification dialog.
Navigation: Left-click and drag to rotate, scroll the mouse wheel to zoom, and click/drag the mouse wheel button to pan. Alternatively, the arrow keys can be used to pan, and Shift combined with up/down arrow keys to zoom.
Point Size Display: Use Ctrl + and Ctrl - to adjust the display size of the points.
Preview Window: Click and drag the top of the Live Classification Preview window to move, resize, or dock the window somewhere else.

Classification and Extraction Shared Settings

Resolution - The resolution at which to analyze the point cloud. This distance, specified in linear units or a multiple of the layer point spacing, will determine the local neighborhood size used for any given point when completing this analysis unless there is a specific overriding parameter for the method. For example, resolution is not used by the Grid MCC ground classification, the smoothing SVM pole classification, or the powerlines classification. These have their own algorithm specific bin size/resolution.

Classification

The classification of points can provide valuable intelligence to a point cloud. These built in tools are designed to identify features in point clouds based on attributes and structures in the point cloud that typically match a type of structure, such as ground, building, etc.

Use the available settings to tailor the tools to best fit your point cloud and the structure of the features you are looking to classify. The majority of settings will be available in the classification section, and a few additional options are also available in the corresponding section of the Classification and Extraction Shared Settings section. Some classification tools have multiple method options as well.

Multiple classifications can be selected at the same time to automatically run one after another. When multiple classifications are selected, they will run in a specific order as follows: noise, ground, power lines, building, vegetation, pole, then any custom classes or subclasses. Additionally, if a Height above Ground filter is set for noise classification, then noise classification will be run a second time after ground is classified to take the newly-identified ground points into account. For buildings, vegetation, and poles, any grid method classifications will be run before max likelihood methods.

Automatic Ground Classification - Prerequisite for all other classifications.

Building Classification - Identifies buildings and flat plane features.

Powerline Classification - Identifies cable/line features.

Vegetation Classification - - Identifies vegetation points, with optional distinctions in tree height.

Pole Classification - Identifies cylindrical pole-like objects.

Noise Classification- Removing outlying noise points from a cloud improves the quality of the cloud and the accuracy of all other classifications.

Tree Trunk Classification - Identify points representing tree trunks of a user-specified size.

Enable Report

Enable this option to save a text summary of the classification settings and results to the Reports folder, and export a JSON settings file for future reuse or sharing.
To load the saved settings .json file: Open the automated Point Cloud Analysis tool, check the box to Enable Custom Feature Models, then from the list of options that appear, choose Load Models.

Report information

Class Changes: How many points of each classification category were reassigned a different classification.
Input : Input point cloud layers and how many points were processed.
Classifier Setup: Classification settings as entered into the Automatic Point Cloud Analysis dialog.
Report Artifacts: File path to the saved report and json model.

The following report items pertain to segmentation-based (Max-Likelihood) classification methods, and may or may not appear, depending on the results of the classification:

Processing: Models run and any tiling of the processing
Classification Health: Provides information about the results. For example, "Healthy" is a successful classification, versus "Critical" can indicate no points were assigned and suggestions for improving results. "Notable" can indicate that segments best matched a class that was not enabled in this run.
Unclassified reasons: When a segment doesn't get a class, the report explains why:
- LowConf (Low Confidence): The segment best matched that class, but its probability score fell below the classifier's required confidence threshold (e.g., segments looked like Medium Vegetation, but were not clear enough to officially classify).
- NotAssign (Not Assigned / Recipe Disabled): The segment matched a class, but that class was not enabled in your current run settings (e.g., segments matched Pole, but you only ran the Building classification recipe).
- Other: The segment was excluded due to geometric, height, or spatial filtering criteria (such as falling outside specified area or footprint limits).
Per-class confidence: Shows the distribution of the classifier's top prediction confidence scores across each class, grouped into 10% percentage brackets (from <10% up to >=90%). Here, p1 refers to the highest probability score calculated for a segment's best-matching class.
Preferred class: Indicates which class the remaining unclassified segments are likely to be. Even though these segments failed to earn an official label (due to low confidence scores, disabled recipes, or geometric rules), the algorithm tracks which class was their top candidate.
Assigned with low confidence: Identifies point segments that were successfully assigned to a class, but where the classifier's top choice barely beat out the second-best candidate.
Columnar drop-reason breakdown: Summary explaining why potential vertical, column-like objects (such as utility poles) were rejected by the classifier rather than saved as confirmed features.

Advanced options

Geometric Segmentation - Identifies features in the point cloud based on user chosen attributes and thresholds. A dynamic and hands on method for Lidar classification.

Use Custom Segmentation Parameters - check this box to further tailor the segmentation based Automatic Classification methods (Max Likelihood) using the setting currently chosen in the Geometric Segmentation tool.
For example, when classifying ground points with the Max Likelihood method, you can open the Geometric segmentation tool and change the curvature value to better match the data's terrain.

Extraction

Features to Extract - Extract representational point, line, area, and mesh features (with relevant attributes) from classified point clouds.

Flags to Apply

Model Key-Point Identification - Find and apply a flag to model key-points required for generating a TIN surface.

Key Concepts for using the Automatic Point Cloud Analysis tool in Global Mapper

Resolution or Bin Size

Most critical parameter in Global Mapper's automatic point cloud classification
Defines the size of the local neighborhood used for analysis.
Smaller resolution = higher detail, slower processing.
Larger resolution = could result less detail, but faster processing.
Two ways to set resolution:
- By point spacings:
  - Ideal for low-density point clouds (e.g., historical fixed-wing lidar, point clouds with less than 3 points per square meter).
  - A good starting point is 3 to 5 point spacings.
  - Ensures enough points (at least 3) per neighborhood.
- By ground units (e.g., meters or feet):
  - Ideal for higher-density point clouds (e.g., from UAVs or terrestrial scanners).
  - Prevents impossibly high-resolution processing.
  - A good starting point is around 1 meter, adjustable based on feature size.

Classification Methods

Grid method:
- Originally designed for low-density fixed-wing aerial lidar.
- Analyzes points within grid cells.
- Faster processing but may require more user input to set parameters.
Max Likelihood method:
- Modern and powerful tool for high-density point clouds.
- Uses segmentation to group points with similar characteristics.
- Assigns a unique segment ID to each segment for quality assurance and easier manual cleanup.

Further Considerations

Global Mapper's automatic classification organizes data into categories like noise, ground, and non-ground.
You may need to process data files or certain areas individually (with the Specify Bounds option) if terrain varies greatly.
Filter out noise points first (e.g., high/low elevation values) to improve the accuracy of subsequent classifications.
A standard workflow is: classify noise first, then ground, which improves the accuracy of all other classifications.
Experimenting with settings? Check the Enable Report option under Classification to save to see a report of the classification results, including how many points changed classification. The classification settings will also be saved as a .json file for future use or for easy sharing with other Global Mapper users.

About Point Cloud Classification

Global Mapper's Lidar auto-classification tools allow you to identify and classify noise, ground and non-ground points from unclassified point clouds. When you run auto classifications on multiple input files, the same classification parameters are applied to each input file. If ground conditions and quality of Lidar points vary by file, then files should be classified separately. The accuracy of auto-classification results will depend on the level of detail and the quality of point cloud data, in addition to the operator's knowledge of ground/terrain conditions. Post-processing and Quality Assurance steps can be used to optimize classification settings when defaults do not yield satisfactory results. Global Mapper's Path/Profile tool has an available Lidar Toolbar, so that manual classification of points in a profile view may be conducted.

Outlying elevation values can lead to spurious or inaccurate classification results. Laser pulse returns from obstructions such as haze, air craft, or birds or multiple reflections from tree canopies can result in outlying points with very high elevation values. Ground structures and trees or tree canopies can create multiple reflections, leading to excessively long travel times back to the Lidar sensor and creating points with outlying low elevation values. These points with anomalous high and low elevation values are noise points, and can be most easily and quickly seen by coloring or drawing point clouds by elevation. The noise points will expand the expected elevation range for the area. If your Lidar data has noise points, and you plan on using the Auto-Classify Ground Points and Auto-Classify Non-Ground Points tools, you will want to Auto-Classify Noise points first. You can also filter out extreme spikes by using the option to Delete Samples Over X Standard Deviations from the Mean in the Lidar Load Options menu.

Most Lidar data will contain a certain percentage of ground points, along with a number that are unclassified. The Auto-Classification Ground Points tool can be used to identify previously unclassified ground points for use in eliminating unclassified points as possible building or tree features, or for employing in the creation of a digital terrain model. Once ground points have been classified a different algorithm may be employed to identify previously unclassified non-ground features, such as buildings, trees, and powerlines by using the Auto-Classify Non-Ground Points. In the classification of non-ground points, relatively flat surfaces that are above the height determined to be ground height will be classified as buildings and those that are vertically offset from neighboring points by user-defined parameters will be classified as trees or powerlines.

The typical workflow for auto-classification of unclassified Lidar points is to classify noise points, ground points, and then non-ground points. Identifying previously unclassified noise points will improve the ground auto-classification results, and classifying previously unclassified ground points will improve the auto-classification of non-ground points. Some of the classification dialogs and will allow for user defined bin size settings, the bin size you want to set is based on the resolution of your Lidar data.