基于GPU并行优化的网格参数化算法  

Mesh parameterization based on GPU parallel optimization

作　　者：吴璇 张举勇[1] Wu Xuan;Zhang Juyong(School of Mathematical Sciences,University of Science and Technology of China,Hefei 230026,China)

机构地区：[1]中国科学技术大学数学科学学院,安徽合肥230026

出　　处：《信息技术与网络安全》2020年第9期16-23,共8页Information Technology and Network Security

摘　　要：网格参数化是计算机图形学、数字几何处理领域的研究热点,在动画、医疗、工业设计等领域中都发挥着重要作用。现有参数化方法主要思路是构造一个高度非线性的全局优化问题,因此计算效率低,难以并行。提出了一种可并行、可扩展的参数化算法。该算法通过引入辅助变量。然后使用交替方向乘子算法(Alternating Direction Method of Multipliers,ADMM),迭代优化每个面和每条边上的子问题得到参数化映射。为了验证算法模型的高效性,使用GPU加速,相比于现存单线程算法,本文算法因为高度并行化运行时间缩短了至少百倍以上。Mesh parameterization is a research hotspot in the field of computer graphics and digital geometry processing.It plays an important role in animation,medical treatment,industrial design and other fields.Existing methods formulate this problem as a global optimization problem.Due to its high nonlinearity and global optimization of the model,it is very difficult to solve efficiently and parallelize.This paper presents a parallel and scalable algorithm for mesh parameterization.The proposed method solves this problem by introducing a set of auxiliary variables.Then using ADMM(Alternating Direction Method of Multipliers),this problem can be easily solved by optimizing small problems for each face and each edge iteratively.To verify the efficiency of the proposed method,we implement the proposed algorithm via GPU,reduce the running time by at least 100 times compared with single thread implementation due to the high parallelism of the proposed algorithm.

关 键 词：并行计算 交替方法乘子法 网格参数化 优化算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]