🟠 Extract Point Cloud Based on Detection Results

Function Description

This operator is a key node connecting 2D vision detection with 3D point cloud processing. It precisely extracts 3D point clouds of regions where each detection target is located from a corresponding ordered point cloud based on input 2D detection results (such as masks or bounding boxes output by "Object Detection" operators).

Usage Scenarios

After completing object detection on 2D images, this operator can be used to quickly segment out 3D point clouds corresponding to each target for subsequent 3D analysis. For example, by extracting point clouds of specific objects, background and irrelevant objects can be effectively removed, providing target point clouds for robot grasping.

Input Output

Input

Detection results: A list containing multiple detection instances, each containing mask and other information.

Point cloud: Ordered point cloud whose dimensions must be consistent with the 2D image that generated the detection results.

Output

Point cloud: A point cloud list where each point cloud element corresponds to one detection instance in the input.

Parameter Description

This operator has two versions:

  • Extract Point Cloud Based on Detection Results: Processes point clouds without normal information.

  • Extract Point Cloud Based on Detection Results (with normals): Processes point clouds with normal information. Both have identical core functionality and parameters, differing only in the type of point cloud data they process.

Minimum Bounding Box

Parameter Description

Extracts point clouds based on the minimum oriented bounding boxes of detection instances. Suitable for scenarios where object shapes are relatively regular, capable of enclosing targets with minimal rectangles.

Parameter Adjustment

  • Bounding box scaling ratio: Used to adjust the size of the extraction region. A value of 1 means using the original bounding box size, greater than 1.0 means expanding the bounding box (e.g., 1.1 means expanding by 10%), less than 1.0 means shrinking the bounding box (e.g., 0.9 means shrinking by 10%).

  • Enable reverse: Reverses extraction logic. Disabled means extracting point clouds inside the specified shape; enabled can be used to remove identified objects and retain background.

Mask

Parameter Description

Extracts point clouds based on pixel-level masks of detection instances. High precision, capable of precisely extracting point clouds of arbitrarily shaped objects, but with relatively larger computational load.

Parameter Adjustment

  • Extract mask edges only: Used to analyze object contour features. When enabled, the operator only extracts point clouds at mask contour edges, not the entire mask interior.

  • Edge thickness: Sets the pixel width of edge contours. The larger the value, the thicker the extracted edge point cloud band, containing more points.

  • Enable reverse: Reverses extraction logic. Disabled means extracting point clouds inside the specified mask; enabled can be used to remove identified objects and retain background.

Min Max Values

Parameter Description

Extracts point clouds based on axis-aligned bounding boxes (horizontal and vertical rectangles) of detection instances. Fastest computation speed, suitable for scenarios where precision requirements for extraction regions are not high.

Parameter Adjustment

  • Bounding box scaling ratio: Used to adjust the size of the extraction region. A value of 1 means using the original bounding box size, greater than 1.0 means expanding the bounding box (e.g., 1.1 means expanding by 10%), less than 1.0 means shrinking the bounding box (e.g., 0.9 means shrinking by 10%).

  • Enable reverse: Reverses extraction logic, defaults to disabled, when enabled can reduce extraction range.

Central Circular Region

Parameter Description

This method is mainly used to extract central point clouds for calculating point cloud center positions and normals.

Parameter Adjustment

Circle radius ratio: Defines the radius of the circular region, which is a proportion of the short side length of the detection instance bounding box. The larger the value, the larger the extracted circular region. If you want to obtain the core region of an object, you can use a smaller value; if you want to cover most of the object, increase this value.