基于区间树硬件加速索引的Marching Cubes算法  

Hardware-Accelerated Interval Tree Index for Marching Cubes

作　　者：张迎平[1,2] 高国贤[1,2] 陆一峰[1,2] 朱登明[1] 王兆其[1] 

机构地区：[1]中国科学院计算技术研究所虚拟现实实验室,北京100190 [2]中国科学院研究生院,北京100049

出　　处：《计算机辅助设计与图形学学报》2012年第7期871-878,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61173067);国家自然科学基金-广东联合基金(U0935003)

摘　　要：为了充分利用GPU的海量线程并行架构,提高等值面可视化效率,提出一种基于区间树硬件加速索引的Marching Cubes算法.该算法在预计算阶段利用GPU构造多区域的区间树作为体数据体素的值域索引;在实时运行阶段根据用户给定的阈值,通过该索引并行地搜索活跃体素,并生成活跃体素的多级索引,然后分配线程处理活跃体素,抽取并绘制等值面.将文中算法应用到不同体数据上的实验结果表明,其能够显著地提高现有Marching Cubes算法的效率;与现有的GPU基准算法相比,最高能达到4～10倍的加速比.This paper proposes a novel hardware-accelerated interval tree index for marching cubes, which exploits the massively parallel architecture of modern GPU for accelerating isosurface extraction. In the pre-computation stage, our method partitions volume data into sub-volumes and then indexes voxels based on its range interval in each sub-volume totally on the GPU. In the runtime stage, this index is used to parallelly locate and process active voxels which are intersected by the given isosurface value in marching cubes. We illustrate our method by using various volume data, in which it outperforms all other known GPU-based marching cubes. The computation speed of our method increases up to 4-10 times in comparison with existing GPU benchmark algorithms.

关 键 词：等值面 移动立方体算法 区间树 活跃体素 GPU 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]