Hole Inpainting Algorithm for Half-Organized Point Cloud Obtained by Structured-Light Section System

Abstract

The advances in sensors and data processing technologies enrich the types of 3D point clouds acquirement, empowering numerous extensive and novel applications such as 3D reconstruction in various scenarios. However, the hole defects affect the accuracy and fidelity of the acquired point clouds, hindering further development and application of 3D point clouds. Aiming at the hole defects in point clouds, a Bayesian hole inpainting algorithm for the half-organized point cloud is proposed, where the point cloud is obtained by a structured-light section system. The algorithm establishes a Bayesian probability model in the hole region, which adopts specific distributions of the point cloud to estimate the maximum likelihood parameter. Simulation and experimental results show that the proposed approach outperforms other competing algorithms significantly in repairing various types of holes, both in objective and subjective qualities. In addition, the proposed algorithm has better scalabilities in the cases of wrong topology definition and self-intersection confusion. This is the first algorithm specially designed for hole inpainting in half-organized point clouds, which maximizes the comprehensive consideration of local features and global optimization, supplemented by targeted prior knowledge of density, Riemannian manifold, and discrete attributes.