Abstract: Concerning the problems of low efficiency and less processing points of the traditional algorithm for point cloud simplification, according to the slicing principle in the rapid prototyping with feature-preserving and low computational complexity, a parallel slicing algorithm was designed and implemented for more than ten millions point cloud of Light Detection And Ranging (LiDAR) data. The point cloud model was layed with the slicing principle and every layer was sorted according to the angle. Incorporating the parallel computation framework of Compute Unified Device Architecture (CUDA) proposed by NVIDA and taking the highly parallel performance advantages of the programmable Graphics Processing Unit (GPU), and parallel execution of the single slice point cloud simplification with the multi-thread of GPU was done, which improved the algorithm efficiency. Finally, a comparing experiment was done with three groups of point cloud data in different order of magnitudes. The experimental results show that the efficiency of the proposed algorithm has 1-2 order of magnitude higher than that of traditional algorithm under the condition of keeping the model characteristics and not changing the compression ratio.

Key words: massive point cloud, simplification, slicing algorithm, Compute Unified Device Architecture (CUDA), Graphics Processing Unit (GPU), parallel computing

摘要： 针对传统点云简化算法效率低且处理点数少的缺陷，结合快速成型领域的切片原理顾及特征计算复杂度低的特点，设计并实现了适合千万级海量激光雷达（LiDAR）点云的并行切片简化算法。该算法根据切片原理对点云模型分层并按照角度排序，利用NVIDA的统一计算设备架构（CUDA）和可编程图形处理器（GPU）高度并行的性能优势，使用GPU多线程高效并行地执行单层切片点云简化，提高了算法效率。最后，应用3组不同数量级点云模型分别进行简化对比实验。实验结果表明：在保持模型特征与压缩比不变的情况下，所提算法效率高出传统基于CPU的串行切片算法1～2个量级。

关键词: 海量点云, 简化, 切片法, 计算设备架构, 图形处理器, 并行计算