Point Cloud Processing

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 combine multiple point clouds to reconstruct a 3-D scene, or build a map with 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.

To perform point cloud registration, the process of aligning two or more point clouds to a single coordinate system, you typically start with one point cloud as the reference, or fixed point cloud, and then align other, or moving, point clouds to it. The absolute pose of a point cloud refers to its global position and orientation with respect to a reference frame, often known as the world coordinate frame. Computer Vision Toolbox provides various registration techniques to register a moving point cloud to a fixed point cloud. These techniques include iterative closest point (ICP), normal distributions transform (NDT), phase correlation, and coherent point drift (CPD). You can also use the Lidar Registration Analyzer (Lidar Toolbox) app to interactively register and compare the results of using different registration techniques, tuning parameters, and preprocessing steps.