基于CUDA的场面全景图像柱面投影并行算法  

Parallel Implementation of Airport Panoramic Image Cylindrical Projection Based on CUDA

作　　者：陈朝勇[1] 王凯[2] 吴敏[2] 姚辉[2] 张翔[3] CHEN Chao-yong;WANG Kai;WU Min;YAO Hui;ZHANG Xiang(Company of Air Control,The 2nd Research Institute,Civil Aviation Administration ofChina,Chengdu 610041,China;Research and Development Center,The 2nd Research Institute,Civil Aviation Administration ofChina,Chengdu 610041,China;School of Electronic Engineering,University of Electronic Science and Technology of China,Chengdu 611731,China)

机构地区：[1]中国民用航空总局第二研究所空管公司,成都610041 [2]中国民用航空总局第二研究所科研中心,成都610041 [3]电子科技大学电子工程学院,成都611731

出　　处：《控制工程》2020年第9期1636-1641,共6页Control Engineering of China

基　　金：国家自然科学基金项目(U1733111,U1433112);民航空管科技项目(2018XZ-214);四川省科技计划项目(2017JZ0007)。

摘　　要：在机场宽场景下全景视频图像拼接中,柱面全景图生成和显示相对比较简单,因而被广泛采用,但存在视频分辨率及焦距的增加处理耗时不断增长的问题。研究并设计标准运算平台(Compute Unified Device Architecture,CUDA)中视觉处理器(Graphics Processing Unit,GPU)的并行计算原理,提出一种基于CUDA的场面全景图像柱面投影并行算法,将数据格式对齐、设备数组与二维纹理绑定以及柱面投影坐标计算等功能模块利用并行策略转移至GPU中进行加速处理。实验结果表明,优化后的算法能明显提升处理效率,并根据视频分辨率和焦距的提高,最大加速比提升约9倍。The generation and display of cylindrical panoramic image is relatively simple,which is widely used in airport panoramic image,but the computation load is very huge especially for large and remote video image.In this paper,an efficient approach was proposed to optimize cylindrical projection algorithm on visual processor(Graphics Processing Unit,GPU)via standard computing platform(Compute Unified Device Architecture,CUDA),and time consuming modules,such as data format aligning,device array and two-dimensional texture bounding,cylindrical projection coordinate calculating,were implemented in GPU.The experimental results show the optimized algorithm can reduce the time consumption significantly.As the image size and distance increase,it can get a more than 9 x speedup.

关 键 词：柱面全景图 柱面投影 标准运行平台 视觉处理器 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]