Robust Plane Extraction using Supplementary Expansion for Low-Density Point Cloud Data

Abstract

Robust plane extraction from point cloud is important for 3D environment modeling in autonomous navigation and 3D object manipulation in robotics. Conventional plane extraction approaches using repetitive decomposing and merging process, however, suffered from low accuracy when the point cloud data density is low or varies significantly. In this paper, a fast and robust plane extraction algorithm is introduced by proposing an expansion stage after every decomposition stage unlike traditional decompose-and-merge approaches that continue to decompose until a terminal condition is reached. The proposed method uses the Mahalanobis distance from the center of the plane for plane expansion while previous works utilized the orthogonal distance in the process of plane extension. This enables the algorithm to omit points that are orthogonally close to the plane but do not actually belong on the plane. Various experimental results show that the proposed structure leads to more accurate and succinct results under the conditions where traditional decomposing and merging algorithms fall behind in performance. The number of divided planes is reduced by 73% and this shortened the elapsed time by 62%. In the end, the proposed method excelled in performance successfully where point cloud density falls low or where different planes meet to make an edge.