半全局匹配算法的多基线扩展及GPU并行处理方法  

Study on semi-global matching algorithm extended for multi baseline matching and parallel processing method based on GPU

作　　者：刘昭华[1] 杨玉霞[1] 马大喜[1] 杨靖宇 

机构地区：[1]江西理工大学建筑与测绘工程学院 [2]61733部队

出　　处：《测绘科学》2014年第11期99-103,共5页Science of Surveying and Mapping

基　　金：江西省教育厅科学技术研究项目(GJJ11474)

摘　　要：文章分析了半全局匹配算法基本原理,将其扩展应用于遥感影像的多基线匹配,在进一步提高匹配可靠性的同时,保留了算法的规则结构;并研究其图形图像处理器(GPU)细粒度并行处理技术,重点探讨匹配代价立方体生成与聚合过程的核函数优化策略与线程组织方案,最后利用Tesla C2050GPU并行加速卡对3幅UCD航空影像进行MVLL多基线匹配半全局优化GPU并行处理实验证明了该算法的有效性和高效性。This paper firstly analyzed the principle of semi-global matching algorithm, then extended it into multi baseline matching to improve the matching reliability and keep the regular processing struc- ture, and then realized its fine granularity parallel processing technologies based on GPU, in which espe- cially studied the kernel function optimization strategies and the GPLI threads＇ executing scheme of matc- hing cost cube computing and aggregating; at last did a system experiment to MVLL＇ s semi-global opti- mize algorithm using three UCD aerial images based on Tesla C2050 GPU card to improve its effectiveness and productiveness.

关 键 词：半全局匹配 多基线匹配 动态规划法 图形图像处理器 并行处理 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]