Point Cloud Processing

Preprocess, visualize, register, fit geometrical shapes, build maps, implement SLAM algorithms, and use deep learning with 3-D point clouds

A point cloud is a set of data points in 3-D space. The points together represent a 3-D shape or object. Each point in the data set is represented by an x, y, and z geometric coordinate. Point clouds provide a means of assembling a large number of single spatial measurements into a dataset that can be represented as a describable object. Point cloud processing is used in robot navigation and perception, depth estimation, stereo vision, visual registration, and in advanced driver assistance systems (ADAS). Computer Vision Toolbox™ algorithms provide point cloud processing functionality for downsampling, denoising, and transforming point clouds. The toolbox also provides point cloud registration, geometrical shape fitting to 3-D point clouds, and the ability to read, write, store, display, and compare point clouds. You can also combine multiple point clouds to reconstruct a 3-D scene.

You can use pcregistericp, pcregisterndt, pcregistercorr, and pcregistercpd to register a moving point cloud to a fixed point cloud. These registration algorithms are based on the Iterative Closest Point (ICP) algorithm, the Normal-Distributions Transform (NDT) algorithm, the phase correlation algorithm, and the Coherent Point Drift (CPD) algorithm, respectively. You can build a map with the registered point clouds, detect loop closures, optimize the map to correct for drift, and perform localization in the prebuilt map. For more details, see Implement Point Cloud SLAM in MATLAB.

Figure showing a point cloud of two concentric point clouds combined, a sensor angle computation, and a point cloud representing a teapot

Functions

expand all

Read and Write Point Clouds

`pcread`	Read 3-D point cloud from PLY or PCD file
`pcwrite`	Write 3-D point cloud to PLY or PCD file
`pcfromdepth`	Convert depth image to point cloud
`pcfromkinect`	Point cloud from Kinect for Windows
`velodyneFileReader`	Read point cloud data from Velodyne PCAP file

Store Point Clouds

`pcviewset`	Manage data for point cloud based visual odometry and SLAM
`pointCloud`	Object for storing 3-D point cloud

Visualize Point Clouds

`pcviewer`	Visualize and inspect large 3-D point cloud
`pcshow`	Plot 3-D point cloud
`pcshowpair`	Visualize difference between two point clouds
`pcplayer`	Visualize streaming 3-D point cloud data
`showShape`	Display shapes on image, video, or point cloud

Process Point Clouds

Preprocess

`pcbin`	Spatially bin point cloud points
`pcdenoise`	Remove noise from 3-D point cloud
`pcdownsample`	Downsample a 3-D point cloud
`pcnormals`	Estimate normals for point cloud

Find and Remove Points

`findPointsInCylinder`	Find points within a cylindrical region in a point cloud
`findPointsInROI`	Find points within a region of interest in the point cloud
`findNearestNeighbors`	Find nearest neighbors of a point in point cloud
`findNeighborsInRadius`	Find neighbors within a radius of a point in the point cloud
`removeInvalidPoints`	Remove invalid points from point cloud

Segment Point Clouds

`pcsegdist`	Segment point cloud into clusters based on Euclidean distance
`segmentGroundFromLidarData`	Segment ground points from organized lidar data
`segmentLidarData`	Segment organized 3-D range data into clusters
`pcbin`	Spatially bin point cloud points

Register Point Clouds and Create Maps

Register Point Clouds

`pcregistercorr`	Register two point clouds using phase correlation
`pcregistericp`	Register two point clouds using ICP algorithm
`pcregistercpd`	Register two point clouds using CPD algorithm
`pcregisterndt`	Register two point clouds using NDT algorithm

Transform Point Clouds

`rigidtform3d`	3-D rigid geometric transformation
`pctransform`	Transform 3-D point cloud

Align or Combine Point Clouds

`pcalign`	Align array of point clouds
`pccat`	Concatenate 3-D point cloud array
`pcmerge`	Merge two 3-D point clouds

Determine Loop Closure Candidates

`findPose`	Localize point cloud within map using normal distributions transform (NDT) algorithm
`scanContextDistance`	Distance between scan context descriptors
`scanContextDescriptor`	Extract scan context descriptor from point cloud
`scanContextLoopDetector`	Detect loop closures using scan context descriptors

Optimize Poses

`createPoseGraph`	Create pose graph
`optimizePoses`	Optimize absolute poses using relative pose constraints

Create Localization Map

pcmapndt Localization map based on normal distributions transform (NDT)

Fit Point Clouds to Geometric Models

`pcfitcylinder`	Fit cylinder to 3-D point cloud
`pcfitplane`	Fit plane to 3-D point cloud
`pcfitsphere`	Fit sphere to 3-D point cloud
`pcnormals`	Estimate normals for point cloud
`fitPolynomialRANSAC`	Fit polynomial to points using RANSAC
`ransac`	Fit model to noisy data
`cylinderModel`	Parametric cylinder model
`planeModel`	Object for storing parametric plane model
`sphereModel`	Object for storing a parametric sphere model

Blocks

expand all

Visualize Point Clouds

Point Cloud Viewer

Visualize streaming point cloud data sequence

Topics

Choose a Point Cloud Viewer
Compare visualization functions.
Choose SLAM Workflow Based on Sensor Data
Choose the right simultaneous localization and mapping (SLAM) workflow and find topics, examples, and supported features.
Implement Point Cloud SLAM in MATLAB
Understand point cloud registration and mapping workflow.
The PLY Format
The Stanford Triangle Format
Getting Started with Point Clouds Using Deep Learning
Understand how to use point clouds for deep learning.
Choose Function to Visualize Detected Objects
Compare visualization functions.
Labeling, Segmentation, and Detection (Lidar Toolbox)
Label, segment, detect, and track objects in point cloud data using deep learning and geometric algorithms

Featured Examples

Build a Map from Lidar Data Using SLAM

Process lidar data to build a map and estimate a vehicle trajectory using simultaneous localization and mapping.

Open Script

Point Cloud Classification Using PointNet Deep Learning

Train PointNet for point cloud classification.

Open Live Script

Build a Map from Lidar Data

Process 3-D lidar sensor data to progressively build a map, with assistance from inertial measurement unit (IMU) readings.

Open Script

Detect and Track Vehicles Using Lidar Data

Track vehicles using measurements from a lidar sensor mounted on top of an ego vehicle.

Open Script

Train Classification Network to Classify Object in 3-D Point Cloud

Train a classification network to classify objects in a 3-D point cloud.

Open Live Script

3-D Point Cloud Registration and Stitching

Combine multiple point clouds to reconstruct a 3-D scene using Iterative Closest Point (ICP) algorithm.

Open Live Script

Design Lidar SLAM Algorithm Using Unreal Engine Simulation Environment

Record synthetic sensor data from a 3-D simulation environment, and develop a simultaneous localization and mapping (SLAM) algorithm using the recorded data.

Open Script

Ground Plane and Obstacle Detection Using Lidar

Process 3-D lidar data from a sensor mounted on a vehicle by segmenting the ground plane and finding nearby obstacles.

Open Script