Merging-based hierarchical octree management and visualization for large-scale engineering point clouds
[Objective]With the continuous advancement of intelligent water conservancy,3 D laser scanning technology has at-tracted significant attention due to its comprehensive monitoring capabilities,rapid scanning speed,and high precision.Howev-er,these point clouds,characterized by massive scale,unstructured nature,and uneven density,present formidable challenges in engineering data management and visualization.Aiming at the limitations of the"top-down,step-by-step subdividing"point cloud indexing construction method,which are not well-suited for massive point cloud chunking or multi-point cloud input scenar-ios,a"bottom-up,small-to-large"octree construction method for massive point data management and fast point cloud rendering was proposed.[Methods]First,a global voxel-based indexing and encoding conversion system is established based on the global information of point clouds.This framework facilitates the rapid octree indexing for massive point cloud data through operations such as point-by-point allocation and grid merging,which are executed in parallel by multi-threaded processing.Furthermore,a non-redundant sampling strategy is adopted to build the LOD(level of details)models,coupled with multi-threaded dynamic scheduling technology,to achieve high-quality visualization and rendering of massive point clouds.[Results]Experimental result demonstrated that the merge-based construction strategy could handle point cloud data exceeding 5 billion points.The peak mem-ory consumption during the indexing process was 2.78 GB,and the processing efficiency reached 2 million points per second.Compared to point cloud processing tools like Point Cloud Library(PCL)and CloudCompare,the proposed method exhibited superior performance in memory management and indexing struct construction efficiency,as well as point cloud rendering.[Conclusion]The management and visualization of large-scale engineering point cloud data are fundamental to advancing the full-process research of engineering point cloud monitoring.The key lies in optimizing the rapid construction of spatial indexes for ultra-large point cloud datasets.The"subdivide first,merge later"processing strategy effectively alleviates spatial overlap con-flicts in chunk data,with clear algorithm logic,making it particularly suitable for index structures based on spatial subdivision mappings such as quadtrees and octrees.
3D laser scanningmassive point cloudoctreeinternal and external memory dynamic schedulingmerge algo-rithmlevel of detailswater conservancydigital twin
万方数据
维普
