Point Cloud Organization

Function Description

Utilizes Camera internal reference to convert the input, usually unordered, camera coordinate system Pointcloud into an Ordered Point Cloud.

The converted Pointcloud forms a two-dimensional grid in structure, and the three-dimensional points in the original Pointcloud will be mapped to the corresponding positions of this two-dimensional grid.

If some positions in the grid do not have corresponding three-dimensional points in the original Pointcloud, these positions will be marked as invalid values (NaN).

Usage Scenarios

  • Align with 2D Image : Associate 3D Pointcloud data with 2D Image captured by the same camera from the same perspective at the pixel level. For example, find the corresponding depth value or 3D coordinates for each pixel of the Image, or assign Image colors to the Pointcloud.

  • Generate Depth Map : The Z-coordinates (depth information) in the Ordered Point Cloud can be directly extracted or transformed to generate a depth map aligned with the reference Image.

  • Apply 2D Image Processing Algorithms : Certain algorithms designed for 2D Image, such as filtering and feature extraction, can be applied to the transformed data structure after Pointcloud organization (e.g., applying 2D mean filtering to the depth values of the Ordered Point Cloud).

Input/Output

Input Items

Image: Reference Image.

Camera Coordinate System Pointcloud: A list of one or more Pointclouds.

Camera internal reference: In-camera reference matrix.

Output Items

Organized Pointcloud: Pointcloud processed through organization. If the input is a Point Cloud List, the output will also be a list of organized Pointclouds corresponding to each input Pointcloud.

Parameter Description

This operator has two versions:

  • Point Cloud Organization : Processes Pointcloud without normal information.

  • Point Cloud Organization (with Normals) : Processes With Normal Point Cloud information.

Both have identical core functions and parameters, differing only in the type of Pointcloud data processed.

Merge input point clouds

Parameter Description

Controls whether to first Merge input point clouds into a single, large Pointcloud when the input is a Point Cloud List, before performing the organization process.

Parameter Tuning Guide

  • Off (default): If the input Pointcloud is a list containing multiple Pointclouds, the operator will independently organize each Pointcloud. The output Ordered Point Cloud will also be a list, containing the corresponding number of organized Pointclouds. Suitable for situations where multiple independent Pointclouds need to be organized separately.

  • Turn on node functionality: If the input Pointcloud is a Point Cloud List, the operator will first combine all Pointcloud data in this list into a single Pointcloud. Then, it will perform the organization operation on this merged Pointcloud. The output will be a list containing only one Ordered Point Cloud. Suitable when multiple Pointcloud segments of a scene are intended to be treated as a whole to generate a unified ordered view.

Align Image Dimensions

Parameter Description

This parameter controls whether the 2D dimensions (width and height) of the output Ordered Point Cloud strictly match the input Image dimensions.

Parameter Tuning Guide

  • Off (default): The dimensions of the output Ordered Point Cloud will be determined by the coverage area after projecting the actual Pointcloud data onto the image plane using Camera internal reference. This means the width and height of the output Pointcloud may not be identical to the original dimensions of the input Image, but it will ensure all effectively projectable points are contained within the most compact 2D grid possible. [cite: 2, 6]

  • Turn on node functionality: The width and height of the output Ordered Point Cloud will strictly match the input Image dimensions. In this case, if some 3D points project outside the Image range, these points will not appear in the output Ordered Point Cloud. Suitable for scenarios where the generated Ordered Point Cloud (e.g., a depth map) needs to correspond precisely pixel-for-pixel with the original Image. [cite: 2, 6]