计算机工程与应用 ›› 2009, Vol. 45 ›› Issue (10): 160-162.DOI: 10.3778/j.issn.1002-8331.2009.10.048

• 图形、图像、模式识别 • 上一篇    下一篇

实时的改进互信息匹配及其GPU实现

崔世林,田 斐   

  1. 南阳理工学院 电子系,河南 南阳 473004
  • 收稿日期:2008-10-09 修回日期:2008-12-24 出版日期:2009-04-01 发布日期:2009-04-01
  • 通讯作者: 崔世林

Improved real-time mutual information matching and its implemented on GPU

CUI Shi-lin,TIAN Fei   

  1. Department of Electronic,Nanyan Institute of Science and Technology,Nanyang,Henan 473004,China
  • Received:2008-10-09 Revised:2008-12-24 Online:2009-04-01 Published:2009-04-01
  • Contact: CUI Shi-lin

摘要: 为提高模板匹配的速度和精度,提出了一种基于GPU实现的改进的实时互信息配准方法。为了算法不至于太复杂而不能在GPU下运行,使GPU临时变量减少到4个,然后利用GPU逐像素计算模板和焊缝图像的互信息。将所提算法应用到焊缝跟踪系统中,通过GPU来执行简化后的互信息计算,实验表明,所提出的算法可以完成每秒约30帧的图像焊缝跟踪。

Abstract: To improve the rate and accuracy of model matching,this paper proposes a new algorithm of real-time module matching based on mutual-information which is implemented on GPU.To reduce complex of the algorithm to compliant the GPU,the bicolour process is used to reduce the temp variables to four.Experimental of tracking weld seam system shows that the proposed algorithm can accomplish 30 frames per second.