摘要
提出一种基于统一计算设备架构(CUDA)加速的尺度不变特征变换(SIFT)快速计算方法,用以解决SIFT特征提取计算过程耗时过长的问题.该方法充分利用图像处理单元(GPU)在并行计算、浮点计算、内存管理等方面的优势,合理分配主机端和设备端的资源及其在SIFT特征计算中所承担的角色.实验表明,与CPU架构下的SIFT特征提取算法相比,本文算法可以大幅度加快SIFT特征提取的计算速度,其加速比随着SIFT特征点数目的增加而增加,在本文实验中最大加速比可达19.54.
A novel algorithm for accelerating scale-invariant feature transform（SIFT） was presented on the basis of compute unified device architecture（CUDA）,which could solve the time-consuming problem in SIFT feature extracting.This algorithm took the advantages of graphic processing unit（GPU） in parallel computation,float point computation and memory management,and reasonably allocated the computational resources and the corresponding computational tasks to the host and device in the SIFT feature extracting.Experimental results show that,compared with the CPU-based acceleration algorithm,the proposed CUDA-based algorithm greatly speeds up the extracting of SIFT features.The acceleration ratio increases with the number of SIFT feature points.The maximum acceleration ratio in the experiments was 19.54.
出处
《东北大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2013年第2期200-204,共5页
Journal of Northeastern University(Natural Science)
基金
国家自然科学基金资助项目(61101057)
