Modeling of large-scale point model

This paper proposes efficient data structures for point-based rendering and a real-time and high quality rendering algorithm for large-scale point models. As a preprocessing, large-scale point model is subdivided into multiple blocks and a hierarchical structure with Minimal Bounding Box (MBB) property is built for each block. A 3D R-tree index is constructed by those MBB properties. A linear binary tree is created in every block data. During rendering, the model is deal with block by block. Fast view-frustum detection based on respective MBB and 3D R-tree index are first performed to determine invisible data blocks. For visibility detection, this project proposes three algorithms which are back point visibility detection, view point-dependent visibility detection and depth-dependent visibility detection. Visible blocks are then rendered by choosing appropriate rendering model and view point-dependent level-of-detail. For determined level-of-detail, corresponding point geometry is accessed from the 3D R-tree and the linear binary tree (K-D tree). Adaptive distance-dependent rendering is accomplished to select point geometry, yielding better performance without loss of quality. The experiment system is developed in C# program language and CSOpenGL 3D graphic library. The point-cloud data sampled from several great halls of Forbidden City are used in experiment. Experimental results show that our approach can not only design to allow easy access to point data stored in Oracle databases, but also realize real-time rendering for huge datasets in consumer PCs. Those are the grounds for the modeling and computer simulation with point-cloud data.