Vitreous: 3D Point Cloud Processing Skills

INFO

Vitreous is a module in the Telekinesis SDK containing skills for 3D point cloud processing.

It provides a comprehensive set of skills for 6D pose estimation, 3D object detection, point cloud filtering, segmentation, and clustering. Vitreous is designed for building applications such as vision-based pick-and-place, bin picking, and robot navigation, where reliable 3D perception is critical.

When to Use Vitreous?

Use Vitreous when your application relies on 3D perception with point clouds, such as:

• Vision-based pick and place
• Bin picking and object localization
• Robot navigation and environment understanding
• Scene reconstruction and spatial reasoning

What does Vitreous Provide?

Vitreous includes a collection of modular skills for:

• 6D pose estimation
• 3D object detection
• Point cloud filtering and downsampling
• Segmentation and clustering
• Geometric feature extraction

How to Use Vitreous?

To use the skills from Vitreous, simply use:

from telekinesis import vitreous

Here is a minimal example:

from telekinesis import vitreous

from telekinesis import vitreous
pcd = vitreous.load_point_cloud("scene.ply")
segments = vitreous.segment_point_cloud(pcd)

Overview of Skills

SDK Call	Description
calculate_axis_aligned_bounding_box	Computes the axis-aligned bounding box (AABB) of a 3D point cloud.
calculate_oriented_bounding_box	Computes the oriented bounding box (OBB) of a point cloud aligned with principal axes.
calculate_plane_normal	Estimates the unit normal vector from plane coefficients.
calculate_point_cloud_centroid	Calculates the centroid (geometric center) of a 3D point cloud.
calculate_points_in_point_cloud	Counts the number of points in a point cloud.
cluster_point_cloud_using_dbscan	Applies DBSCAN clustering to identify dense regions in 3D point clouds.
cluster_point_cloud_based_on_density_jump	Splits a point cloud at the steepest density discontinuity using the density jump splitting algorithm.
convert_mesh_to_point_cloud	Converts a triangle mesh to a point cloud using sampling methods.
create_cylinder_mesh	Creates a 3D cylinder mesh aligned along the Z-axis and centered at the origin.
create_plane_mesh	Creates a plane mesh whose center overlaps with the global origin.
create_sphere_mesh	Creates a sphere mesh whose center overlaps with the global origin.
create_torus_mesh	Creates a 3D torus mesh aligned along the Z-axis and centered at the origin.
estimate_principal_axis_within_radius	Estimates a principal axis from a local point cloud neighborhood using PCA.
estimate_principal_axes	Calculates the principal axes of a 3D point cloud using PCA or OBB method.
filter_point_cloud_using_bounding_box	Extracts points within a specified 3D axis-aligned bounding box.
filter_point_cloud_using_cylinder_base_removal	Extracts one or both circular base points of a cylindrical structure.
filter_point_cloud_using_mask	Filters a point cloud using a boolean mask.
filter_point_cloud_using_oriented_bounding_box	Extracts a region of interest using an oriented 3D bounding box.
filter_point_cloud_using_plane_defined_by_point_normal_proximity	Extracts points near a plane defined by a point and normal vector.
filter_point_cloud_using_plane_proximity	Extracts points near a plane defined in general form.
filter_point_cloud_using_plane_splitting	Filters a point cloud by keeping only points on one side of a plane.
filter_point_cloud_using_radius_outlier_removal	Removes isolated points based on neighbor radius.
filter_point_cloud_using_statistical_outlier_removal	Removes noise by analyzing point-to-neighbor distance distributions.
filter_point_cloud_using_uniform_downsampling	Reduces point cloud density by selecting every nth point.
filter_point_cloud_using_viewpoint_visibility	Removes points not visible from a specific viewpoint.
filter_point_cloud_using_voxel_downsampling	Reduces point cloud density by averaging points within voxel cells.
add_point_clouds	Adds two point clouds together by combining their points.
subtract_point_clouds	Subtracts the target point cloud from the source point cloud.
scale_point_cloud	Scales a point cloud about a specified center point.
apply_transform_to_point_cloud	Transforms a point cloud using a rigid body (Euclidean) transformation.
project_point_cloud_to_plane	Projects a point cloud orthogonally onto a plane defined in general form.
project_point_cloud_to_plane_defined_by_point_normal	Projects a point cloud onto a plane defined by a point and normal vector.
reconstruct_mesh_using_convex_hull	Performs convex hull reconstruction on a given point cloud.
reconstruct_mesh_using_poisson	Poisson surface reconstruction for creating smooth surfaces from point clouds.
register_point_clouds_using_cuboid_translation_sampler_icp	Performs ICP with translation samples uniformly distributed in a cuboid region.
register_point_clouds_using_fast_global_registration	Performs fast global registration using FPFH features without RANSAC.
register_point_clouds_using_point_to_plane_icp	Performs point-to-plane ICP registration with optional robust kernels.
register_point_clouds_using_point_to_point_icp	Performs point-to-point ICP (Iterative Closest Point) registration.
register_point_clouds_using_rotation_sampler_icp	Performs ICP with multiple orientation initializations sampled on a sphere.
segment_point_cloud_using_color	Segments points based on color similarity to a target color.
segment_point_cloud_using_plane	Segments planes from point clouds using RANSAC algorithm.