Research on Segmentation of Pear Shape from Unorganized Point Clouds

Abstract

With the advent of 3D scanner, accurate segmentation of 3D fruit shape from unorganized point clouds has been turned out to be the most challenging task in scientists and engineers in reverse engineering. This paper herein proposes efficient and robust approach to extract pear shape from background. At first, an interactive, non-local denoising algorithm is employed to efficient denoise the pear scans; Second, geometric properties, including normal, variation and curvature, are estimated by covariance analysis; Third, a Recursive Region Increment (RRI) is proposed to add the geometric similarity points to a base set, to generate an ultimate set only including the points of pear shape; Forth, point clouds is linearized for rapidly rending in the post processing. Finally, segmentation algorithm applied on ten range scans of a pear demonstrates that our algorithm reduces the number of pear point clouds by 88.3%, proves the validity and practicability of this method in pear segmentation.