摘要
针对3D模型海量点云数据存在的空间冗余问题,提出一种基于TSLVQ(tree structure lattice vector quantization)的静态点云有损渲染算法。算法旨在利用层级嵌套网格的集合,解决渲染低效的问题。首先对整个点云进行包围盒封装,多层量化,把数个较小尺度的截断包围盒嵌入到一个较高尺度的截断包围盒单元中,每一步量化过程采用8叉树方法将包围盒分割为八个最佳尺寸的空或非空小包围盒;最后在最高深度的层级里,用包围盒来代替整个小包围盒中全部的点。同时,算法可自行设定8叉树的深度,从而任意控制编码的复杂度和精度,满足渲染的实时性要求。实验结果表明,与现有的网格有损压缩算法相比,提出的算法能在保证模型重建精度的基础上具有较好的空间分解优势,实现实时渲染效果。
Concerning the problems that in point cloud with a lot of temporal redundancy, this paper presented a novel lossy compression approach based on tree-structure lattice vector quantization for static point cloud. The proposed approach utilized the hierarchical packing of embed truncated lattices, which could solve the problem of representation inefficiency. Firstly, it enclosed the point cloud with a cube, then truncated the point cloud, embedded smaller-scale cubes into the bigger-scale cube, the multi-stage procedure of quantization, in every level, this method divided the cube to 8 sub-cutes or voxels as empty or non-empty subspace. Finally, in the predefined max level, it used the cube to represent all the other points in the same cube. Furthermore, it allowed for predefining max level to control coding complexity and coding precision to meet the real-time rendering requirement. Experimental results show that the octree decomposition has obvious advantages in 3D object real-time representation comparing with the existing gridding lossy compression, guaranteeing 3D model reconstruction precision.
作者
石祖旭
曾安
Vincent Ricordel
Nicolas Normand
Shi Zuxu;Zeng An(College of Computer, Guangdong University of Technology, Guangzhou 511400, China;LS2N Laboratory, University de Nantes, Nantes 44306, France)
出处
《计算机应用研究》
CSCD
北大核心
2019年第7期2205-2209,共5页
Application Research of Computers
基金
国家自然科学基金资助项目(61772143,61300107,61672168)
广州市科技计划资助项目(201601010034,201804010278)
广东省大数据分析与处理重点实验室开放基金资助项目(201801)
关键词
向量量化
树型结构网格编码矢量量化
有损压缩
3D点云
vector quantization
tree structure lattice vector quantization(TSLVQ)
lossy compression
3D point cloud
