文档介绍：摘要计算机视觉是当前的研究热点,而双目立体视觉作为计算机视觉的一个分支, 越来越得到重视和关注,其应用也越来越广泛,如无人导航、工业控制、三维测量和虚拟现实等。立体匹配技术是双目立体视觉的关键技术之一,其基本思想是: 在相关的约束规则下,按照一定的代价函数寻找不同视角的两幅或者多幄图像之间的像素点的对应关系。本文以立体匹配算法及其基本理论的研究为基础,重点探讨了非参量的特征区域匹配和基于不变量的立体匹配技术,进而实现了基于上述匹配的三维重建算法。首先,系统地介绍了立体匹配的一般原理、关键技术,以及在三维重建中需要的标定原理。在基于区域的匹配技术中引入权值矩阵,通过实验证明权值矩阵可提高匹配可靠性;其次,针对图像获取存在干扰以及在标定过程中存在标定误差的问题,引入了基于非参量的特征区域匹配技术和基于不变量的图像匹配方法, 详细介绍了这两种方法的原理,实现了基于这两种方法的立体匹配算法。实验表明,这两方法都具有较强的健壮性和较好的抗干扰性,且后者还具有较快的速度, 无论是在特征提取还是匹配都比最近流行的SIFT(Scale InvariantFeature Transforml 快很多;最后,在以上工作的基础上,进行了三维重建理论和方法的研究,实现了一个三维重建系统。关键词:图像匹配权值矩阵三维重建非参量 puter vision isthe current research focus,and binocular stereo vision,a branch puter vision,has increasingly gained attention stereo vision-based applications are increasingly being wildly used such asdrone navigation, industrialcontrol,three—dimensional measurement,virtual reality and etc·Stereo matching techn0109y is one of thekeytechnologies of stereo vision,the basic idea Bv thecertain cost function andrelevantmatching restrictions,it isof essence toget the corresDondent points from images which are captured from two or more different viewpoints. The paper has realized and focused matching and the invariant based stereo on thenon-parameter characteri sticregion matching technique on the basis of stereo matching algorithm and ,the paper systematically deals wltn tne binocular stereo match’sgeneral principle,the key technologies,and the calibration Drinciple which needed inthe weight matrix hasbeen introduced to matching technique based on region and it has been proved more reliable and emcient;Secondly,the paper proposes thenon-parameter characteristic re910nmatching technique and the invariant stereo match algorithms owing tointerference in1mage acauisition and existing calibration experiment indicates that these two methods are strong inrobustness and good inanti-j amming pared with recentlypopular SIFT method,the invariant stereo matchalgo