摘要
针对畸变差改正算法的处理速度不高和CUDA实现算法加速的设备局限性问题,提出了一种OpenCL并行改进畸变差纠正算法实现加速的方法。该方法是对传统的畸变差纠正算法进行并行改进,通过调用计算机GPU的计算单元实现算法加速;采用CPU+GPU的异构模式实现算法加速,将传统算法中逐像素密集计算部分分配到GPU进行处理;与CUDA实现算法加速针对NVIDIA显卡设备不同,OpenCL并行改进的算法没有了设备的限制。实验结果表明,相对于传统算法来说,影像畸变差纠正处理速度显著提升,总体加速比最高达5.976,计算部分加速比最高达到63.432,同时在AMD显卡设备上也得到了较好的加速效果。
In order to improve the processing speed of the distortion correction algorithm and solve the problem of the device limitation caused by the CUDA algorithm acceleration,an OpenCL parallel improved distortion correction algorithm is proposed to achieve acceleration.The method is to improve the traditional distortion correction algorithm in parallel,and to accelerate the algorithm by calling the computing unit of the computer GPU.Using CPU+GPU heterogeneous mode to achieve algorithm acceleration,the pixel-intensive computing part of the traditional algorithm is allocated to the GPU for processing.Accelerating algorithms with CUDA Different from NVIDIA graphics devices,the OpenCL parallel improved algorithm has no device limitations.The experimental results show that:compared with the traditional algorithm,the speed of image distortion correction processing is significantly improved,the overall acceleration ratio is up to 5.976,the calculation part of the acceleration ratio is up to 63.432,and on the AMD graphics device it also has better accelerated effect.
作者
于梦华
王双亭
李英成
朱祥娥
刘晓龙
YU Menghua;WANG Shuangting;LI Yingcheng;ZHU Xiange;LIU Xiaolong(Henan Polytechnic University,Jiaozuo,Henan 454000,China;Key Laboratory for Aerial Remole Sensing Technology of National Administration of Surving,Mapping and Geoinformation (NASG),Beijing 100039,China;TopRS Technology Co.,Lte,Beijig 100039 China)
出处
《遥感信息》
CSCD
北大核心
2019年第3期88-92,共5页
Remote Sensing Information
基金
国家重点研发计划项目(2017YFB0503004)
关键词
OPENCL
算法加速
畸变差改正
并行改进
加速比
OpenCL
algorithm acceleration
distortion correction
parallel improvement
speedup
