基于OpenCL的加速鲁棒特征算法并行实现

doi:10.3969/j.issn.0253-2778.2017.10.002

中国科学技术大学学报 ›› 2017, Vol. 47 ›› Issue (10): 808-816.DOI: 10.3969/j.issn.0253-2778.2017.10.002

基于OpenCL的加速鲁棒特征算法并行实现

郭景，陈贤富

中国科学技术大学信息科学技术学院，安徽合肥 230027

收稿日期:2017-04-01 修回日期:2017-08-05 出版日期:2017-10-31 发布日期:2017-10-31
通讯作者: 陈贤富
作者简介:郭景，男，1992年生，硕士生.研究方向:图像匹配、目标跟踪.E-mail：guojing6@mail.ustc.edu.cn

Parallel implementation of surf algorithm based on OpenCL

GUO Jing, CHEN Xianfu

School of Information Science and Technology, University of Science and Technology of China, Hefei 230027, China）

Received:2017-04-01 Revised:2017-08-05 Online:2017-10-31 Published:2017-10-31

摘要/Abstract

摘要： 加速鲁棒特征算法(speed up robust features, SURF)的时间复杂度大，传统串行计算的方法，实时性难以保证.针对上述问题，提出一种基于OpenCL架构的SURF并行实现方法.首先对算法中的积分图的计算、Hessian响应图、特征点主方向、特征点描述等步骤实施数据并行和任务并行处理，并给出详细的算法流程.接着从OpenCL架构的数据传输、内存访问以及负载均衡等方面优化算法性能.实验结果表明，该算法对不同分辨率的图片均实现了10倍以上的加速比，一些高分辨率的图片甚至可以达到39.5倍，并且算法适用于多种通用计算平台.

关键词: 加速鲁棒特征, 开放运算语言, 图像处理器, 并行计算

Abstract: SURF algorithm has high computational complexity and can not meet the real-time requirement. To solve these problems, a parallel SURF algorithm based on OpenCL was presented. Firstly, data parallelism and task parallelism model were applied to the calculations of the integral images, Hessian detector, orientation and descriptor, and the detailed algorithm process was given. Secondly, the performance of the parallel algorithm was optimized from data transmission, memory access and load balancing. The experimental results show that the algorithm can achieve more than 10 times speedup for images with different resolution, and some high-resolution images can even reach up to 39.5 times. Furthermore, it can be applied to a variety of general purpose computing platforms.

Key words: SURF, OpenCL, GPU, parallel computing

中图分类号:

TP391

郭景，陈贤富. 基于OpenCL的加速鲁棒特征算法并行实现[J]. 中国科学技术大学学报, 2017, 47(10): 808-816.

GUO Jing, CHEN Xianfu. Parallel implementation of surf algorithm based on OpenCL[J]. Journal of University of Science and Technology of China, 2017, 47(10): 808-816.

参考文献

［1］
赵向阳, 杜利民. 一种全自动稳健的图像拼接融合算法[J]. 中国图象图形学报: A 辑, 2004, 9(4): 417-422.
ZHAO Xiangyang, DU Limin. An automatic robust mosaic algorithm[J]. Journal of Image and Graphics, 2004, 9(4): 417-422.
[2] WANG Y T, FENG Y C. Data association and map management for robot SLAM using local invariant features[C]// International Conference on Mechatronics and Automation. Takamatsu, Japan: IEEE, 2013: 1102-1107.
[3] BAY H, TUYTELAARS T, VAN GOOL L. Surf: Speeded up robust features[J]. Computer Vision & Image Understanding, 2006, 110(3): 404-417.
[4] LEUTENEGGER S, CHLI M, SIEGWART R Y. BRISK: Binary robust invariant scalable key points[C]// Proceedings of the International Conference on Computer Vision. Washington: IEEE, 2011: 2548-2555.
[5] 赵春阳, 赵怀慈. SURF 算法并行优化及硬件实现[J]. 计算机辅助设计与图形学学报, 2015, 27(2): 256-263.
ZHAO Chunyang, ZHAO Huaici. Parallel optimized method and hardware implementation of SURF algorithm[J]. Journal of Computer-Aided Design & Computer Graphics, 2015, 27(2): 256-263.
[6] SVAB J, KRAJNIK T, FAIGL J, et al. FPGA based speeded up robust features[C]// Proceedings of the International Conference on Technologies for Practical Robot Applications. Woburn, USA: IEEE, 2009: 35-41.
[7] 刘金硕, 曾秋梅, 邹斌, 等. 快速鲁棒特征算法的 CUDA 加速优化[J]. 计算机科学, 2014, 41(4): 24-27.
LIU Jinshuo, ZENG Qiumei, ZOU Bin, et al. Speed-up robust feature image registration algorithm based on CUDA[J]. Computer Science, 2014, 41(4): 24-27.
[8] 徐晶, 曾苗祥, 许炜. 基于 GPU 的图片特征提取与检测[J]. 计算机科学, 2014, 41(7): 157-161.
XU Jing, ZENG Miaoxiang, XU Wei. GPU based image feature extraction and detection[J]. Computer Science, 2014, 41(7): 157-161.
[9] SINHA S N, FRAHM J M, POLLEFEYS M, et al. GPU-based video feature tracking and matching[R]. MSR-TR-2016-19, Workshop on Edge Computing Using New Commodity Architectures, 2006.
[10] CAO J, XIE X F, LIANG J, et al. GPU accelerated target tracking method[J]. Advances in Multimedia, Software Engineering and Computing, 2012, 1: 251-257.
[11] Khronos OpenCL Working Group. The OpenCL Specification Version1.2[EB/OL]. [2017-10-18]https://www.khronos.org/registry/OpenCL/specs/opencl-1.2.pdf.

()
(

[1]	陈锋，张智，李琴剑，陈宇强，陈国良. FCD大数据并行处理的动态任务调度算法[J]. 中国科学技术大学学报, 2018, 48(9): 718-722.
[2]	蔡勇，陈红梅，. MapReduce环境下基于概念分层的概念格并行构造算法[J]. 中国科学技术大学学报, 2018, 48(4): 275-283.
[3]	谢青，彭威，柳瑶阁，黄涛，卢德唐. 改进的储层直接取样随机模拟方法及GPU实现[J]. 中国科学技术大学学报, 2013, 43(8): 626-630.

基于OpenCL的加速鲁棒特征算法并行实现

Parallel implementation of surf algorithm based on OpenCL

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 3

编辑推荐

Metrics

本文评价