大规模天文时序粒子数据的可视化  

Visualization of Large Scale Time-Varying Particles Data from Cosmology

作　　者：单桂华[1] 谢茂金[1] 李逢安 高阳[1] 迟学斌[1] 

机构地区：[1]中国科学院计算机网络信息中心超级计算中心,北京100190

出　　处：《计算机辅助设计与图形学学报》2015年第1期1-8,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(91230115);中国科学院信息化专项项目(XXH2503-02)

摘　　要：时序数据的可视化是理解宇宙结构形成与演化的重要手段.围绕大规模天文数值模拟输出的近百TB粒子时序数据的可视化,针对数据的高动态范围色调映射问题,提出一种基于统计直方图的算法,实现了时序上色调连贯的可视化;同时,在插值重建演化过程时,考虑到模拟输出的每个关键帧的数据依据Hilbert三维填充曲线分布于2048个文件中,在一次可视化中通常有相当部分的文件包含的数据不会进入视锥内,据此提出一种文件尺度上根据前后关键幀预判插值幀可见性的剪裁算法,将前后关键帧可见数据文件的序号集合作为插值幁可见数据文件的序号集合;对裁剪结果进行实时插值和投影,通过裁剪算法大幅降低计算量、存储和I/O,并通过Hilbert哈希元胞快速完成裁剪过程;最后给出了算法的性能和效果分析.可视化结果表明文中算法可以直观、有效地表达大规模数据所包含的宇宙结构形成细节与演化信息.Time-varying data visualization is a fundamental way of understanding the formation and evolutionprocess of the cosmology structures. In our terascale time-varying cosmology data visualization, wepresent a statistic based tone mapping algorithm for the data with extreme high dynamic range both in spatialand time dimensions, with which we gained novel tone-coherent visualization results. At the mean time,when reconstructing the evolution process, as the data in one time step are distributed on 2048 files in Hilbertspace filling curve way, we found that quite a part of the data files are invisible in most visualization,thus we proposed a visibility culling algorithm for the interpolated data based on the nearest key frame pair.Our algorithm puts the union set of visible file IDs as the culling result, through which we dramatically reducethe computation, memory consumption and I/O operations. The culling process is efficiently done byusing the Hilbert hash cell. Last, we analyze the performance and image quality .Our visualization resultsshow that algorithms proposed here are great help for efficiently expressing forming detail of the cosmologystructure, and also great help for gaining insight into the evolution of the Universe.

关 键 词：时序数据 高动态范围 粒子绘制 科学可视化 色调映射 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]