An adaptive approach for primitive shape extraction from point clouds

Abstract

In the paper, an adaptive approach for primitive shape extraction from point clouds is presented. The approach extends RANSAC segmentation algorithm in two ways: adaptive primitive shape detection based on histogram analysis of points’ deviations from their corresponding fitted primitive shapes; getting boundary points of the fitted primitive shapes and trimming uniform points of primitive shapes according to boundaries. The data structure of a fitted primitive shape contains two parts: parameter vectors that define the primitive shape; boundary points of the primitive shape. By placing uniform points on primitive shapes and trimming away the points outside of boundaries, we got trimmed uniform points each fitted primitive shapes. Afterwards, we got mesh models of primitive shapes by exerting Delaunay algorithm on each set of trimmed uniform points. At last, we output the final model by calibrating primitive shapes’ positions and orientations and trimming primitive shapes according to their intersection lines.