Page 20 - 中国仿真学会通讯2020第1期
P. 20
PARALLEL INTERVISIBILITY ALGORITHM
ON GPU
Liankai Yao1,2, Gang Guo1.3
1 Innovation Research Center, Beijing Huaru Science and Technology LTD
2 yaolk@ icloud.com
3 hndzgg@ aliyun.com
The intervisibility of long⁃distance lines of sight must consider the influences of the earth
curvature and the precision of terrain data. Otherwise, there are large errors to the
computational results. A new algorithm has been developed to work out these problems by
dividing a line of sight into several segments in different GCS tiles and computing the
intersections with geometrical curved surfaces instead of sampling. In this paper, this
intervisibility algorithm is implemented on GPU to further improve the performance.
Experiments show that the performance speedup of 10 times is achieved.
Keywords: Intervisibility algorithm; GPU parallel computing; CUDA implementation.
1. Introduction the new algorithm6 which has the advantages of
wide application range and high accuracy has
Intervisibility algorithm plays an important been developed based on Geographic Coordinate
role for simulation, modeling, communication and System ( GCS ) . Long⁃distance LoS which spans
scientific computation1,2,3 to determine whether a several GCS tiles is divided into several segments
line of sight( LoS) is intervisible and where is the in different GCS tiles to eliminate the influence of
occlusion point. The importance and difficulty of the earth curvature. In addition, the results are
intervisibility algorithm are embodied in got by determining whether the segments intersect
calculating results accurately and rapidly with the terrain curved surfaces constructed by
considering the influences of high⁃precision bilinear interpolation.
terrain data and the earth curvature.
High⁃precision terrain data is stored in
The implementation of intervisiblity algorithm Regular Square Grid ( RSG ) model. The
using traditional methods4,5 such as Janus intervisibility computing in every RSG is massive,
method, Dyntacs method, Modsaf method and repetitive and relatively independent. For this
Bresenham method had problems with the kind of compute⁃intensive algorithm, CUDA
accuracy of computational results inevitably. ( Compute Unified Device Architecture) has the
Firstly, the influence of the earth curvature was better performance7 to implement parallel
not taken into account especially for long⁃distance computing on NVIDIA General Purpose GPU.
LoS. Secondly, the computational performance
would decrease with improving precision of terrain The structure of this paper is as follows.
data. Finally, occlusion points were positioned Section 2 introduces the computational steps of
just by sampling. intervisibility algorithm based on GCS tiles.
Section 3 introduces the parallel implementation
To make up the above⁃mentioned shortages, on GPU. Computational results and performance
speedups are got by some numerical experiments
17