基于CUDA的并行改良随机抽样一致性算法  

Parallel Improved RANSAC Based on CUDA

作　　者：苗青[1] 付忠良[1] 赵向辉[1] 徐可佳[1] 

机构地区：[1]中国科学院成都计算机应用研究所,四川成都610041

出　　处：《四川大学学报（工程科学版）》2010年第4期111-116,135,共7页Journal of Sichuan University (Engineering Science Edition)

基　　金：四川省科技支撑计划资助项目(2008SZ0100)

摘　　要：针对传统RANSAC的许多局限性——样本多、模型复杂或数据错误率高时计算效率低,模型检验精度与数据错误率不易合理设置,无法批处理同模型不同样本集,提出一种基于CUDA的RANSAC并行改良,在保证计算结果置信概率与传统RANSAC一致的前提下,同时对抽样、解模型及检验模型并行同步处理,最终选择出符合要求的最优模型参数。以NVIDIAGPU支持的CUDA为并行计算环境,挖掘其硬件架构的通用计算特性,设计并实现了RANSAC的高效GPU运算模式。实验表明,改良后的算法能够克服传统RANSAC的诸多局限性,且保留了其简单易用的特点。In order to overcome several disadvantages of traditional RANSAC,a parallel improved RANSAC based on CUDA was presented. With guaranteeing the same confidence of the solution as traditional RANSAC,the proposed method parallelly dosed sampling,computed and verified model,and found out the most suitable parameters. Meanwhile,NVIDIA’s CUDA was chosen as parallel computation environment,and by exploiting the GPU’s hardware feature for general-purpose computing,GPU computational model of RANSAC was designed and implemented. Experiments showed that parallel improved RANSAC based on CUDA can overcome disadvantages of traditional RANSAC,and retain its simplicity and easy-to-use.

关 键 词：并行计算 稳健估计 RANSAC CUDA 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]