October 26, 2021

Point Cloud Segmentation Analysis in Global Mapper Pro

Written by: Mackenzie Mills


With the release of Global Mapper Pro, Blue Marble Geographics continues to develop the point cloud analysis tools in the program. One of the latest additions to the Global Mapper point cloud processing tools is a Segmentation by Spectral Graph Partitioning tool used to identify unique segments of points based on user input parameters. This new functionality falls under the umbrella of automatic classification and can be used for semi-automated classification, but instead of directly applying a class to analyzed points, the segmentation tool applies a segment identification value to each identified cluster of returns.

What is Point Cloud Segmentation?

The Segmentation tool in Global Mapper Pro uses point-to-point similarity measures in order to identify and label unique clusters of points. The characteristics of each point return used in the analysis depend on the available point attributes, and the user input parameters. Based on the similarity between points this tool then breaks the points into discrete segments.

Attribute Evaluation

As the first step of the segmentation analysis, an attribute evaluation is done to determine point-to-point similarity. To customize this section of the analysis, users can choose and weight attribute values that will be considered in the process.

Position – The coordinate location of the point returns is always considered in this analysis. Considering the position ensures the identified segments of points will be contiguous, representing a single feature or clumped together features. Without considering the position, similar points scattered throughout the point cloud could be placed in the same segment providing no meaningful feature identification.

Normal – A point normal is the direction perpendicular to the surface the point represents. To visualize this concept, consider a small neighborhood of points and the surface they represent, then, at a single point location, imagine the point normal as a perpendicular line striking out from the created surface. Considering the point normal value in the similarity analysis helps to identify flat surfaces. If you are looking to identify rood planes or the ground surface, more heavily using the point normal values in the attribute evaluation will likely yield better results.

A screenshot of lidar data to display normal orientation.
The representation of the surface the points create and a point normal is drawn into a cross-section view of a roof.

Intensity – Intensity is the strength of the laser pulse returned to the sensor in the collection of lidar data. The strength of the return is affected by the surface that reflects the laser pulse. The intensity values of a point cloud can be visualized in Global Mapper Pro using the Color Lidar by Intensity draw mode. Considering intensity in the Segmentation analysis helps to create clusters of points that are more likely to be constructed of the same or similar materials.

Return Number – As a laser pulse is sent and returned to a lidar sensor, the pulse can split and bounce multiple times creating multiple returns. This information is collected with the data saved as a return number attribute. Solid surfaces like the ground or buildings are more likely to be composed of single returns, while trees have a structure that allows the pulse to bounce off branches and leaves creating multiple returns. Like intensity, return number can be visualized in Global Mapper Pro with the Lidar Draw Mode options.

Curvature – Curvature refers to the curve created by a local neighborhood of points, similar to how the point normal values consider the shape created by the points. Including curvature in the attribute evaluation considers the consistency of the shape.

While not all these attributes are available for all point clouds, for example photo-derived point clouds do not typically have a return number or intensity, the Segmentation analysis settings allow users to choose and weight the characteristics that should be considered.


After analyzing point similarity, Global Mapper Pro breaks the points into their clusters, or segments based on the determined similarity and user input partitioning values. Some of the segmentation parameters are unitless and more abstract values than Global Mapper users may be familiar with.

Connectivity Threshold – With a wide range of accepted values (0.0001 to 100), the connectivity value is a threshold of algebraic connectivity used to measure the connections between points and cut the data into segments. Overall, a larger value will result in more segments using smaller changes in similarity to split the data up, and a smaller value will result in fewer segments.

Minimum Number of Points in Segment – The minimum number of points value sets the size, in points, of the smallest allowed segment. When setting this parameter, consider the density of the point cloud and the features you are looking to identify to estimate how many points represent a given feature.

Maximum Number of Standard Deviations – A common measure in statistical analysis, standard deviation measures the spread of values from the mean. Setting the maximum number of standard deviations for segmentation narrows the range of point associations to be considered. A larger number of standard deviations create segments that include more weakly connected points, while a smaller number will create segments of more closely connected points.

Maximum Curvature – Maximum curvature sets the allowed amount of curvature within a single segment. If this value is exceeded, the segment will be split into multiple segments.

Curvature as represented by the shape of lidar points.
Points representing a car are split into two segments with the break between them occurring where the curve is most extreme. A higher curvature threshold may help

How can Segmentation analysis be used?

Point cloud segmentation analysis in Global Mapper can be an incredibly versatile tool. Since a classification is not applied to points after executing this tool, it allows you to identify features without explicit classes and/or classify the features after identifying them through segmentation.

Identifying Ground

Using segmentation to identify ground points can work beyond the automatic classification tools as more attributes can be considered. To identify ground in this point cloud, the segmentation tool is used, and position, intensity, and curvature values are considered. After running the Segmentation tool, the dialog box will remain open, and the Lidar Draw Mode will automatically change to Draw Lidar by Segment ID. This process will show the point cloud by the newly created and applied segment ID values. 

Example settings for finding ground points in a point cloud.
With versatile applications, segmentation can be used to identify large or small features in a point cloud.

Using settings that consider a few point attributes and creating segments of many points, 150 or more, that are highly connected, a low connectivity threshold value, segments for ground and large flat roofs identified. 

To classify the identified segments of points as ground, these points can be selected using the Select Lidar Segments tool. Select the first ground segment by clicking on any point within the segment, then hold the Control keyboard key to allow the selection of the second segment. With the ground representing points selected, they can be manually classified as ground. 

A point cloud that has undergone ground point classification.
A clear classification of ground is achieved before moving on to smaller feature identification.

Identifying Cars

Now that a large portion of the points have been identified and classified as ground, these points can be filtered and the focus can be put on smaller features in the dataset. In this particular point cloud, there is a parking lot of cars, and the Segmentation tool in Global Mapper can be used to identify the car features. 

Turning off point classifications to improve processing, with a screenshot of the settings.
Before executing the Segmentation process to identify cars, the ground classified points are filtered so they will not be considered in the analysis.

Using the Filter Lidar Data tool or filtering points within the segmentation dialog box means that only the unclassified points will be considered in the segmentation analysis. Considering point position and curvature, tightening the connectivity threshold, and altering other parameters results in identified segments representing the cars in this point cloud. 

Cars clearly segmented apart from data.
Shown in the Path Profile view, and 3D, segment identified cars are clearly seen.

With points representing cars identified, they can be selected with the Select Segment tool to be classified or deleted. 

The Point Cloud Segmentation tool in Global Mapper provides opportunities to customize the identification of features in order to classify or remove them when processing a point cloud. Combined with the Select Segment tool, segmentation greatly expands the classification capabilities of the program and the efficiency of manual classification. 

If you would try the new Point Cloud Segmentation in Global Mapper Pro, download a 14-day free trial today! If you have any questions, don’t hesitate to get in touch

If you enjoyed this blog, you may also find these other resources useful:

Point Cloud Segmentation Analysis in Global Mapper

Point Cloud Segmentation Analysis in Global Mapper Pro

Companies using Blue Marble’s geospatial technology