文中提出了一种基于GPU加速的精确计算球面距离变换的算法.球面距离变换在球面网格剖分、气象建模等领域有重要的应用.目前主要的球面距离变换算法是基于有序扩散的,即每个球面像素的距离变换都需要参照相邻像素的信息.因而难以通过并行计算加速.在经度-纬度参数域上,该文给出了依次沿纬线和经线独立扫描进行球面距离变换的理论证明,提出了一种新的球面距离变换算法,称为经纬独立扫描算法.同以往的球面距离变换算法相比,该文提出的算法是完全并行的算法,可以在GPU上实现以提高球面距离变换的性能.最后,该文列举了几个快速球面距离变换的应用.%This paper presents an accurate and fast GPU-based spherical distance transform algorithm. Spherical distance transform is very useful in some research areas, such as spherical tessellation, meteorological modeling and so on. At present, the methods in spherical distance transform are mainly based on ordered propagation, in which the distance of every spherical pixel is computed by referencing information of its neighbors. Based on the latitude-longitude parameterizations on a sphere, the authors give proofs which ensure the correctness of executing spherical distance transform by independently scanning each latitude and longitude line in sequence, and propose a novel spherical distance transform algorithm, which called Longitude-Latitude Independent Scanning (LLIS) algorithm. Compared with previous methods, the algorithm can be implemented in parallel on modern GPUs to achieve high performance. Some applications of the fast spherical distance transform are demonstrated at the end of this paper.
展开▼
