3D Preprocessing Function Configuration Guide
About 5252 wordsAbout 18 min
3D preprocessing takes place after point cloud generation and before 3D pose estimation and Pick Point generation. It removes noise, controls point cloud density, extracts geometric features, or prepares data such as normal vectors and contours for later calculations. This page helps you choose functions for issues encountered on site and explains how to adjust and validate their parameters.
Configuration Principles
Identify the current point cloud issue before adding the corresponding function. Adjust only one parameter at a time, and check whether the Target point cloud remains complete, noise is reduced, and subsequent pose or Pick Point results remain stable.
Select a 3D Preprocessing Function
| On-site issue or processing goal | Recommended function | Selection guidance |
|---|---|---|
| Small noise clusters or background point clouds are present | Point Cloud Cluster Denoising | Form point cloud clusters based on the distance between points, then retain the Target point cloud based on each cluster’s point count or position. |
| Isolated noise points are present | Remove Point Cloud Outliers | Use neighborhood statistics to identify and remove outliers that deviate from the main point cloud. |
| The point cloud contains too many points, slowing computation | Point Cloud Downsampling | Reduce the number of points passed to subsequent processing. If only Fine Matching needs to be accelerated, use “Point Cloud Downsampling (Target Point Cloud).” |
| Subsequent processing requires point cloud normal vectors | Calculate Normal Vectors | Calculate a normal vector for each point so subsequent functions can process the point cloud by surface direction. |
| The vertical direction of a tilted sack must be calculated | Calculate Self-Vertical Pose | Provide a pose for subsequent functions that use the object’s own coordinate system. |
| Subsequent calculations require the Target contour | Extract Point Cloud Contour | Extract the outer contour from the instance point cloud. Use this when contour mode is enabled in subsequent processing. |
| The Target and background can be distinguished by color | Filter Point Cloud by HSV Color Range (Hue, Saturation, Value); Filter Point Cloud by Three-Channel Color | Choose according to the color representation used on site, and set upper and lower limits to retain the Target color range. |
| Points outside ROI 3D must be excluded | Select Point Cloud Inside ROI; Optimize Mask by Point Cloud | The first function directly retains points inside ROI 3D; the second uses points inside ROI 3D to correct the instance mask. |
| Points must be removed based on surface direction | Remove Points Above Normal Angle Threshold | Compare the angle between each point cloud normal vector and the specified coordinate axis, and remove points whose angle exceeds the threshold. |
| Obvious planar noise is present | Point Cloud Plane Segmentation | Fit the plane containing the most points, then retain or remove it as needed. |
| Redundant points below the Target or in a specified direction must be removed | Filter Point Cloud Beyond Object Distance | Filter points beyond a distance threshold along the specified axis and in the specified coordinate system. |
| An instance point cloud must be read from a local file | [Expert] Read Instance Point Cloud | Read the point cloud at the specified path. If the path is empty, use the point cloud uploaded in the Target interface. |
3D Preprocessing Functions
Related Documents
Configure the instance point cloud used for 3D preprocessing
Use the preprocessed point cloud for pose matching and estimation
Adjust Pick Points based on the pose and point cloud results