Generating High-Fidelity Texture in RGB-D Reconstruction using Patches Density Regularization. (July 2023)

• Record Type: Journal Article
• Authors: Liu, Xinqi; Li, Jituo; Lu, Guodong
• Is Part Of: Computer aided design. Volume 160(2023)
• Publication Date: 2023-07
• Subjects: RGB-D reconstruction -- Texture generation -- Patches density regularization -- Texture patches fragmentation; Computer-aided design -- Periodicals; Engineering design -- Data processing -- Periodicals; Computer graphics -- Periodicals; Conception technique -- Informatique -- Périodiques; Infographie -- Périodiques; Computer graphics; Engineering design -- Data processing; Periodicals; Electronic journals; 620.00420285
• DOI: 10.1016/j.cad.2023.103516
• Languages: English
• ISSNs: 0010-4485

Abstract

Abstract: We propose a texture-mapping-based online RGB-D reconstruction framework that can generate stable and high-fidelity texture results in near real-time. Different from previous texture generation methods that are prone to output blur results (point blending methods) or require huge computational costs (previous texture mapping methods), our approach achieves high-fidelity and fast texture generation by introducing a new patches density regularization and a series of lightweight optimization methods. The patches density regularization addresses a universal but widely neglected problem: the texture patches fragmentation, which causes discontinuous seams and worsens texture results. By adding this regularization to the optimization process, the fragmentation can be effectively alleviated. To maintain high performance, we design lightweight optimization methods to significantly reduce the computational cost during texture generation. An incremental texture update strategy removes redundant triangular mesh, and a hierarchical-aware seam leveling reduces the number of optimization objects by utilizing light-varying smooth features. Furthermore, the enhanced voxel representation ensures the continuity and stability of the global texture results. Extensive experiments demonstrate that our method achieves a state-of-the-art level, even compared to the mainstream online and offline RGB-D reconstruction frameworks. Graphical abstract: Highlights: A texture mapping-based online RGB-D reconstruction framework to generate high-fidelity texture results. A patches density regularization to solve the texture patches fragmentation problem and improve texture fidelity. An incremental texture update strategy and hierarchical-aware seam leveling to significantly speed up the texture optimization.