Sphere Detection

This algorithm allows detecting specific spheres in a point cloud

Input

A single point cloud.

Output

An annotation representing a sphere.

Description

The algorithm creates a sphere with a specific radius from a given point cloud. It searches for the sphere in a certain Region of Interest (ROI), which can either be the whole point cloud bounding volume or be determined by a Hough-Transform given the sphere’s target radius. Internally it uses a RANSAC approach to find a sphere in the ROI. Following parameters are available:

  • Inlier threshold: The distance in world units until which a point in the RANSAC algorithm is considered to be supporting the current hypothesis.

  • Fixed sphere radius: The radius of the sought-after sphere; set to -1 if not known.

  • Maximal residual error: The maximum residual of supporting points to RANSAC sphere up to which a result is considered valid.

  • Minimum inliers: The minimum number of supporting points in RANSAC down to which a result is considered valid.

  • ROI detection mode: Determines how the ROI is to be determined. Can be either None (in which case the whole bounding volume is used) or Hough Voting, which performs a Hough-Transform on the point cloud to detect the most likely sphere with the given radius.

  • Max voting volume extent: The maximum extent of the volume used for Hough-Transform based ROI detection in world units.

  • Point cloud index skip: The fraction of points to be used for Hough-Transform based ROI detection (determines the increment when iterating through the points, 1 means to use all points).

  • Relative ROI margin: A fraction of the sphere’s radius to be added to the ROI detected by Hough-Transform.