面向OpenCL架构的Harris角点检测算法  

Harris Corner Detection Algorithm on OpenCL Architecture

作　　者：肖汉[1,2] 马歌[2] 周清雷[1] 

机构地区：[1]郑州大学信息工程学院,郑州450001 [2]郑州师范学院信息科学与技术学院,郑州450044

出　　处：《计算机科学》2014年第7期306-309,321,共5页Computer Science

基　　金：国家自然科学基金(41171357);中国博士后科学基金(2012M510176);河南省重点科技攻关项目(132102310003);河南省教育厅科学技术研究重点项目(13A520354)资助

摘　　要：Harris角点检测算法是计算机视觉领域中使用非常广泛的点特征提取算法,它计算简单,稳定性强,但运算速度慢。当前已有算法优化研究一般只针对单一硬件平台,它们很难实现在不同平台上的高效运行。为此提出一种基于开放式计算语言(OpenCL)设计思想的Harris角点检测并行算法,其采用图形处理器(GPU)中共享存储器、常量存储器和锁页内存机制在OpenCL框架下完成影像角点检测的全过程。实验结果表明,基于OpenCL的Harris角点检测并行算法相比CPU上的串行算法可获得的加速比高达77倍,执行效率明显提高,对于大规模数据处理表现出良好的实时处理能力。Harris corner detection algorithm is widely used for extracting feature points in the field of computer vision. It is simple and stable, but inefficient. Currently most of the researches on algorithm optimization are aimed at a single hardware platform,and difficult to achieve the efficient running on different platforms. In this paper, parallel algorithm of Harris corner detection based on the core concept of Open Computing Language （OpenCL） was proposed, so that the whole image corner detection process can be implemented in OpenCL architecture. Finally, implementation of the paral- lel algorithm using mechanism of shared memory and constant memory and pinned host memory in Graphic Processing Unit （GPU） was detailed. The experiments show that the parallel algorithm of Harris corner detection based on OpenCL demonstrates substantial improvement up to 77 times speedup than the serial algorithm running in the CPU, has high efficiency compared with CPU counterpart algorithm, and exhibits great potential for large-scale data process- ing in real-time processing.

关 键 词：图形处理器 开放式计算语言 影像 角点检测 HARRIS算子 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] P237[自动化与计算机技术—计算机科学与技术]