An Improved Adaptive Dynamic LOD Algorithm Based on Large-Scale Point Clouds

Abstract. The rapid development of 3D scanning technology has significantly reduced the cost of acquiring high-precision point cloud data, leading to exponential growth in applications such as digital city modeling, autonomous driving, and virtual reality. However, managing point cloud datasets containing hundreds of millions of points poses severe challenges for storage, transmission, and real-time rendering. Traditional Level of Detail (LOD) techniques struggle to balance efficiency and accuracy, especially under dynamic viewpoints and complex scenes. This paper introduces an enhanced dynamic adaptive LOD algorithm designed to boost the interactivity of large-scale point clouds and eliminate the need for re-computation when data is modified. The research objective is to address the bottleneck in real-time processing of large-scale point cloud data through improved data structures and sampling strategies.Our method builds a multi-resolution data structure that combines octree indexing with spatial sampling to achieve efficient spatial queries. The technical core is a point-voxel hybrid octree based on secondary sampling, which significantly improves visualization efficiency. The innovation lies in the adaptive sampling technique, which dynamically adjusts grid size according to point cloud density, enhancing sampling efficiency and detail precision in both sparse and dense regions.The experimental implementation uses WebGL, Vue, Three.js, Node.js, and MySQL technologies. Test results reveal significant improvements in rendering speed and resource utilization. This research not only enhances real-time rendering capabilities for large-scale point clouds but also has important application value in fields such as GIS and real-time SLAM.