江苏大学学报（自然科学版）2011，Vol.32Issue(3)：336-340,5.DOI:10.3969/j.issn.1671-7775.2011.03.018
基于流形优化法的相机位姿估计
Estimation of camera pose through manifold optimization
摘要
Abstract
To minimize accumulative error of orthogonal iteration in camera tracking, a novel approach based on manifold optimization was proposed to estimate camera pose. Collinear error model was exploited to reformulate pose estimation into minimization of a real-valued function over manifold, and then theorems of differential geometry were adopted for objective optimization. Optimization processes were composed of vector updating and retraction mapping. Zero tangent vector field was searched by Taylor Expansion of objective function in affine tangent space of manifold, and the points which are deviated from manifold were reflected onto mainfold by retraction mapping. The performance of pose estimation by proposed algorithm was compared with that by orthogonal iteration, and its application in augmented reality was given. The results show that the performance of algorithm proposed is better than that of the orthogonal iteration. When noise variance equals l, the relative deviation of new algorithm with synthetical datum is 0.5 ％ for rotation axis and 0.25 ％ for rotation angle, respectively, which are only one fifth and a haft of those for orthogonal iteration.关键词
机器视觉/增强现实/相机跟踪/微分几何/矩阵分解/图像注册Key words
machine vision/ augmented reality/ camera tracking/ differential geometry/ matrix decomposition/ image registration分类
信息技术与安全科学引用本文复制引用
潘绍松,左洪福..基于流形优化法的相机位姿估计[J].江苏大学学报（自然科学版）,2011,32(3):336-340,5.基金项目
国家自然科学基金联合资助项目(60939003/F01) （60939003/F01）
