Leveraging the power of multi-core platforms for large-scale geospatial data processing: Exemplified by generating DEM from massive LiDAR point clouds
Leveraging the power of multi-core platforms for large-scale geospatial data processing: Exemplified by generating DEM from massive LiDAR point clouds
复制标题
DOI：
10.1016/j.cageo.2009.12.008
复制
发表时间：
2010-10-01
影响因子：
4.4
通讯作者：
Wu, Huayi
中科院分区：
文献类型：
--
作者：
Guan, Xuefeng;Wu, Huayi
近年来，激光雷达等空间数据采集技术的进步，导致空间数据量爆炸式增长，对计算能力提出了前所未有的挑战。同时，计算平台的核心CPU，也从单核架构演变为多核架构。这一根本性的变化极大地影响了现有的数据处理算法。本文以大规模机载LiDAR点云生成DEM问题为例，研究了如何利用多核平台进行大规模地理空间数据处理，并展示了多核技术如何提高性能。采用流水线技术开发多核平台的线程级并行性。首先，将原始点云划分为重叠的块。其次，这些离散块被并发地插入到并行管道上。在插值过程中，对中间结果进行排序，最终合并成一个完整的DEM。这种并行化展示了具有高数据吞吐量和低内存占用的多核平台的巨大潜力。这种方法在大大减少处理时间的同时实现了优异的性能加速。例如，在2.0 GHz四核英特尔至强平台上，所提出的并行方法可以在大约12分钟内处理大约10亿个激光雷达点（16.4 GB），并产生27,500 x 30,500栅格DEM，使用不到800 MB的主内存。(C) 2010 Elsevier Ltd.版权所有。
In recent years improvements in spatial data acquisition technologies, such as LiDAR, resulted in an explosive increase in the volume of spatial data, presenting unprecedented challenges for computation capacity. At the same time, the kernel of computing platforms the CPU, also evolved from a single-core to multi-core architecture. This radical change significantly affected existing data processing algorithms. Exemplified by the problem of generating DEM from massive air-borne LiDAR point clouds, this paper studies how to leverage the power of multi-core platforms for large-scale geospatial data processing and demonstrates how multi-core technologies can improve performance. Pipelining is adopted to exploit the thread level parallelism of multi-core platforms. First, raw point clouds are partitioned into overlapped blocks. Second, these discrete blocks are interpolated concurrently on parallel pipelines. On the interpolation run, intermediate results are sorted and finally merged into an integrated DEM. This parallelization demonstrates the great potential of multi-core platforms with high data throughput and low memory footprint. This approach achieves excellent performance speedup with greatly reduced processing time. For example, on a 2.0 GHz Quad-Core Intel Xeon platform, the proposed parallel approach can process approximately one billion LiDAR points (16.4 GB) in about 12 min and produces a 27,500 x 30,500 raster DEM, using less than 800 MB main memory. (C) 2010 Elsevier Ltd. All rights reserved.