一种基于GPGPU的SIFT加速算法  

Speeded-up SIFT on GPGPU

作　　者：杨天天[1] 鲁云萍[2] 张为华 

机构地区：[1]江南大学数字媒体学院,江苏无锡214083 [2]复旦大学并行处理研究所,上海201203

出　　处：《电子技术应用》2015年第1期149-152,160,共5页Application of Electronic Technique

基　　金：国家自然科学基金(61370081)

摘　　要：SIFT是目前应用最广泛的基于局部特征的图像特征提取算法之一,针对其运行速度制约其应用范围的问题,提出在图像处理器(GPGPU)上设计并实现将算法各核心模块映射到GPGPU的计算单元并针对GPUPU特性进行优化的SIFT并行加速算法。测试结果表明,基于GPGPU的SIFT并行算法相比于原始串行版本达到了118.2倍的加速,吞吐量达到了76.86图片/s,相比于已有的技术获得了明显的性能提升。The modem GPGPU is not only a powerful graphic processor but also a highly parallel programmable processor. Stronger arithmetic ability and higher bandwidth bring GPGPU great competitive power in real-time processing systems and high- performance computing systems.This paper designs and implements parallel acceleration algorithms for SIFt on GPGPU. We make best use of GPGPU architectures to accelerate our implementations, including uses of shared memory and texture memory, decreasing allocation and free times of memories on GPGPU and etc. GPGPU usually cooperates with CPU to finish a job. This paper imple-ments SIFT on GPGPU in consideration of the impacts of CPU. Experimental results show this method can achieve a speedup of 118.2X with a throughput of 76.86 images per second.

关 键 词：图像处理器GPGPU SIFT算法 并行 加速 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]