Main Content

segmentLidarData

Segment organized 3-D range data into clusters

collapse all in page

Syntax

labels = segmentLidarData(ptCloud,distThreshold)
labels = segmentLidarData(ptCloud,distThreshold,angleThreshold)
[labels,numClusters] = segmentLidarData(___)
[___] = segmentLidarData(___,NumClusterPoints=[1,Inf])

Description

labels = segmentLidarData(ptCloud,distThreshold) segments organized 3-D range data ptCloud into clusters. The function assigns an integer cluster label to each point in the point cloud and returns the cluster label of all points in labels.

The function groups two neighboring points into the same cluster if their Euclidean distance is less than distThreshold or if the angle between the sensor and two neighboring points is at least 5 degrees.

example

labels = segmentLidarData(ptCloud,distThreshold,angleThreshold) sets the angle constraint for grouping points into the same cluster to angleThreshold.

example

[labels,numClusters] = segmentLidarData(___) also returns the number of clusters.

[___] = segmentLidarData(___,NumClusterPoints=[1,Inf]) also sets the minimum and maximum number of points in each cluster, specified as a 2-element vector or as a scalar value. When you specify NumClusterPoints as a scalar, the maximum number of points in the cluster is unrestricted. The function sets the labels to 0 when clusters are outside of the specified range.

Examples

collapse all

Open Live Script

Create organized synthetic lidar data containing two objects.

ldr = zeros(5,100);
ldr(:,1:50) = 10;
ldr(:,51:end) = 20;
pitch = linspace(-18,18,5);
pitch = repmat(pitch',1,100); 
yaw = linspace(-90,90,100);
yaw = repmat(yaw,5,1);

Convert to Cartesian coordinates.

X = ldr .* cosd(pitch) .* sind(yaw);
Y = ldr .* cosd(pitch) .* cosd(yaw);
Z = ldr .* sind(pitch);
pc = pointCloud(cat(3,X,Y,Z));

figure
pcshow(pc.Location,'r')
title('Unclustered Point Cloud')

Figure contains an axes object. The axes object with title Unclustered Point Cloud contains an object of type scatter.

Set the threshold.

distThreshold = 5;

Segment the lidar data.

labels = segmentLidarData(pc,distThreshold);

Plot the results.

figure
hold on
title('Segmented Clusters')

pc1 = select(pc,find(labels == 1));
pcshow(pc1.Location,'g')

pc2 = select(pc,find(labels == 2));
pcshow(pc2.Location,'y')

Figure contains an axes object. The axes object with title Segmented Clusters contains 2 objects of type scatter.

Open Live Script

Set up the PCAP file reader.

veloReader = velodyneFileReader('lidarData_ConstructionRoad.pcap','HDL32E');

Wait for 0.3 seconds from the beginning of the file, then read the point cloud from the next frame.

veloReader.CurrentTime = veloReader.StartTime + seconds(0.3);
ptCloud = readFrame(veloReader);

Segment and remove the ground plane.

groundPtsIdx = segmentGroundFromLidarData(ptCloud);
ptCloudWithoutGround = select(ptCloud,~groundPtsIdx,'OutputSize','full');

Cluster the remaining points. Distance is in meters.

distThreshold = 0.5;
[labels,numClusters] = segmentLidarData(ptCloudWithoutGround,distThreshold);

Add an additional label for the ground plane.

numClusters = numClusters+1;
labels(groundPtsIdx) = numClusters;

Plot the labeled results. Display the ground plane in black.

labelColorIndex = labels+1;
pcshow(ptCloud.Location,labelColorIndex)
colormap([hsv(numClusters);[0 0 0]])
title('Point Cloud Clusters')

Figure contains an axes object. The axes object with title Point Cloud Clusters contains an object of type scatter.

Input Arguments

collapse all

Point cloud, specified as a pointCloud object. ptCloud is an organized point cloud that stores point coordinates in an M-by-N-by-3 matrix. The points must be organized by pitch and yaw angles in a sequential scanning order, which is typically obtained from laser range finders and the velodyneFileReader.

Distance threshold in world units, specified as a nonnegative scalar. Adjacent points are grouped into the same cluster if the distance between them is less than the distance threshold. To reduce the number of output clusters, increase the value of distThreshold.

Data Types: single | double

Angle threshold, in degrees, specified as a scalar in the range [0, 180]. To reduce the number of output clusters, decrease the value of angleThreshold. The function groups adjacent points into the same cluster if the angle formed by the sensor and the points is greater than the angle threshold. For example, in the figure, the function groups points A and B into the same cluster if the angle formed by the sensor, point A, and point B, is greater than angleThreshold. For more details, see Algorithms.

Data Types: single | double

Output Arguments

collapse all

Cluster labels of all points in the point cloud ptCloud, returned as an M-by-N matrix of integers. Each valid point in ptCloud belongs to a cluster. All points in a cluster are assigned the same integer cluster label, ranging from 1 to numClusters. Invalid points, such as points with Inf or NaN coordinates, are assigned the label 0.

Number of clusters, returned as a positive integer. The number of clusters does not include the cluster corresponding to invalid points and excludes the label value, 0, which is reserved for invalid points.

Algorithms

The segmentLidarData function uses distance and angle thresholds to cluster neighboring points. The function groups two neighboring points into the same cluster if their Euclidean distance is less than the input distThreshold or if the angle between the sensor and neighboring points is greater than or equal to the input angleThreshold. If you do not specify angleThreshold, the function sets this angle to 5 degrees.

For example, suppose angleThreshold is set to 90. Because angles α and β in the figure are both greater than the specified threshold of 90 degrees, the function groups points A, B, and C into the same cluster. Because angle σ is less than the 90-degree threshold, the function groups point D into a separate cluster. Each angle the function uses for clustering is formed by the line from a point to the sensor and the line from that same point to the neighboring point

References

[1] Bogoslavskyi, I. “Efficient Online Segmentation for Sparse 3D Laser Scans.” Journal of Photogrammetry, Remote Sensing and Geoinformation Science. Vol. 85, Issue 1, 2017, pp. 41–52.

Extended Capabilities

C/C++ Code Generation
Generate C and C++ code using MATLAB® Coder™.

Version History

Introduced in R2018a

See Also

| | | |