Design of Point Cloud Data Structures for Efficient Processing of Large-Scale Point Clouds

Existing three-dimensional scanning techniques enable the acquisition of dense point clouds representing the surface of the scanned object. However, the voluminous nature of unordered point cloud data leads to extended data processing times, necessitating the utilization of specific data structures for the management of large-scale point clouds. Addressing the performance degradation issue of prevalent point cloud data structures when dealing with a large quantity of points, this paper initiates a comparative analysis of common point cloud data structures, encompassing grid-based, quadtree, and k-d tree (k-dimensional tree) indexing methods. Through theoretical derivations, an examination of the time complexities of various data structures is undertaken. Building on this theoretical foundation, an empirical quantitative assessment of the real-world performance of distinct data structures is executed. Leveraging the insights gained from these analyses, this paper further capitalizes on the inherent shape characteristics of empirically acquired point cloud data to introduce a novel three-tier hybrid indexed point cloud data structure, accompanied by its corresponding algorithmic functionalities. This innovative structure amalgamates grid-based, quadtree, and k-d tree indexing strategies. Empirical findings demonstrate that, when applied to large-scale point clouds, the proposed three-tier hybrid indexed data structure exhibits enhanced indexing establishment speed and neighborhood search velocity compared to conventional algorithms. Thus, this work establishes a foundational data structure support for subsequent processing and application of large-scale point cloud data.