摘要
无序多视影像的三维重建对噪声非常敏感,错误的匹配关系会影响重建的精度,甚至直接导致重建失败。提出了一种稳健的批处理重建算法,首先利用回路闭合约束剔除可能存在误匹配的三视匹配元,然后以三视匹配元中的三焦张量约束代替传统算法的核线约束来计算所有影像旋转矩阵和相机中心位置的全局最优解。重建过程中引入高效的并查集算法来提取多视匹配点,并利用迭代线性三角形算法计算空间点的三维坐标。实验结果表明,所提算法在重建效率和计算精度方面都能取得较好的结果。
3Dreconstruction of unordered multi-view images is very sensitive to noise.Error matching affects the accuracy of the reconstruction or might even lead to failure.A robust batch reconstruction algorithm is proposed.In the this approach triplets which may contain mismatches are removed using a closed cycle constraint,then the trifocal tensor constraint in triplet is used instead of thetraditional epiploic constraint.Additionally,a linear programming algorithm with the l∞norm is used instead of secondorder cone programming to calculate an global optimum for rotating and locating all the images.An efficientunion finding algorithm is introduced into the reconstruction process to exactly definemulti-view matching points.3Dpoints are computed using iterative linear triangulation.Experimental results show that the proposed method performs satisfactorily in terms of reconstruction efficiency and accuracy.
作者
卢俊
张保明
郭海涛
张宏伟
LU Jun ZHANG Baoming GUO Haitao ZHANG Hongwei(Institute of Geospatial Information, Information Engineering University, Zhengzhou 450052, China)
出处
《武汉大学学报（信息科学版）》
EI
CSCD
北大核心
2017年第1期109-115,共7页
Geomatics and Information Science of Wuhan University
基金
国家973计划(2012CB720000)~~
关键词
三视匹配元
回路闭合约束
批处理重建
并查集
线性规划
image triplet
closed cycle constraint
batch reconstruction
union find
linear program
