计算机工程

• 图形图像处理 • 上一篇    下一篇

基于树形结构的半全局立体匹配算法

葛忠孝,邢帅,夏琴,王栋,侯晓芬,江腾达   

  1. (信息工程大学 地理空间信息学院,河南 郑州 450001)
  • 收稿日期:2015-07-23 出版日期:2016-08-15 发布日期:2016-08-15
  • 作者简介:葛忠孝(1992-),男,硕士研究生,主研方向为图像处理、数字摄影测量;邢帅,副教授、博士;夏琴,硕士;王栋,博士;侯晓芬、江腾达,硕士。
  • 基金项目:
    国家自然科学基金资助项目(41371436);国家“973”计划基金资助项目(2012CB720001)。

Semi-global Stereo Matching Algorithm Based on Tree Structure

GE Zhongxiao,XING Shuai,XIA Qin,WANG Dong,HOU Xiaofen,JIANG Tengda   

  1. (School of Geographic Spatial Information,Information Engineering University,Zhengzhou 450001,China)
  • Received:2015-07-23 Online:2016-08-15 Published:2016-08-15

摘要: 半全局匹配算法中匹配代价的计算不能完全利用待匹配点周围的像素,在影像纹理相似区域会造成误匹配现象。针对该问题,提出一种加入2种树形结构的半全局匹配算法。研究代价函数聚合路径方向对匹配结果的影响,将2种不同的树形结构加入到半全局匹配代价函数中,利用动态规划原理计算新的匹配代价,并从4条路径进行匹配代价的聚合,在视差最大允许范围内求解聚合后的匹配代价,查找出使匹配代价最小的视差作为待匹配点的视差值。实验结果表明,该算法在计算匹配代价时,克服了半全局匹配算法不能充分利用影像像素的问题,结合树形结构的特点可实现对待匹配点周围像素的完全利用,比原有算法具有更高的匹配精度,且将匹配代价聚合的路径由16条减至4条,执行效率提高2倍以上。

关键词: 立体匹配, 半全局匹配, 树形结构, 匹配代价, 动态规划

Abstract: In Semi-global Matching(SGM) algorithm,the pixels surrounding the matching point cannot be fully used in the calculation of matching cost,which always causes mismatching results in the area with similar image texture.To solve this problem,an improved semi-global matching algorithm based on two kinds of tree structures is proposed.The influence of cost formulation aggregation direction is researched.Two different tree structures are added in the cost function of SGM in the proposed algorithm.Then the new matching cost is calculated with dynamic programming and aggregated from 4 paths.In the maximum allowable range of disparity,the disparity of matching point is determined by searching for a value that minimizes the aggregated matching cost.Experimental results demonstrate that the proposed algorithm overcomes the shortcoming of semi-global matching algorithm.Thanks to the characteristics of tree structures,the surrounding pixels can be used completely,which leads to higher accuracy than the original algorithm.In addition,the number of matching cost aggregation paths is reduced from 16 to 4,and the execution efficiency is 2 times higher.

Key words: stereo matching, Semi-global Matching(SGM), tree structure, matching cost, dynamic programming

中图分类号: