Cluster Filter With Normals

Description

Filters the ordered point cloud based on connected component analysis algorithms. It checks the spatial distance between points, considering points close enough as "connected" and clustering them into a block. Then, based on the physical area of the cluster and the number of points, it retains the main object areas and removes fragmented, isolated noise blocks.

Overview

Checks the connectivity between points and filters out small isolated noise blocks based on area/point count thresholds, suitable for removing flying pixels and retaining main objects.

Scenarios

  • Remove Flying Pixel Noise: Depth cameras often produce isolated "flying pixels" at object edges. This operator can effectively remove these unconnected small noise points.

  • Small Region Removal: Removes tiny objects or distractions in the scene that are too small in area to meet detection requirements.

  • Retain Main Body: Quickly extracts major connected regions in the scene with a large number of points and a large area.

Inputs and Outputs

Inputs

Organized point cloud: Organized point cloud data that maintains the 2D structure of camera imaging.

Outputs

Filtered point cloud: Point cloud data retained after connected component filtering.

Parameters

Connectivity Distance Threshold

Description

Defines the maximum spatial distance for two points to be considered "connected/neighbors".

Tuning Instructions

  • Larger value: Looser judgment, connects points slightly further apart, prone to merging broken objects.

  • Smaller value: Stricter judgment, connects only very close points, prone to breaking a single object into multiple fragments.

Parameter Range

[0, 800], Default value: 5

Parameter Range

Values/Range: [0.0, 800.0]

Default: 5.0

Cluster Area Threshold

Description

Defines the minimum physical area required for a connected component to be retained.

Tuning Instructions

If the coverage area of a cluster is less than this value, it is considered noise and removed.

  • Larger value: Stronger filtering, removes larger noise blocks, but may accidentally delete small objects.

  • Smaller value: Retains more details.

Parameter Range

[1, 1000], Default value: 300

Parameter Range

Values/Range: [1, 1000]

Default: 300

Minimum Cluster Points

Description

Defines the minimum number of points required for a connected component to be retained.

Tuning Instructions

If a cluster contains fewer points than this value, it is removed. Usually set to 10-50 to remove a very small amount of isolated noise points.

Parameter Range

[1, 1000], Default value: 10

Parameter Range

Values/Range: [1, 1000]

Default: 10

Maximum Cluster Points

Description

Number of points to judge a valid connected component.

Tuning Instructions

If the number of points in a connected component is greater than this value, the area is considered a valid large object sufficiently large and reliable, and the point cloud in that area is directly retained.

Parameter Range

[1, 1000], Default value: 300

Parameter Range

Values/Range: [1, 1000]

Default: 300

Enable Node

Description

Controls whether this operator performs the filtering operation.

Tuning Instructions

  • Enabled (Default): The operator executes the filtering function normally.

  • Disabled: The operator does not perform any operation and outputs the input data as is.

Parameter Range

Default: True