最小化重投影误差的PFR三维射影重建  

3D reconstruction with minimum reprojection error using PFR

作　　者：徐帆[1] 王宏远[1] 方磊[1] 田文[1] 

机构地区：[1]华中科技大学电子与信息工程系,湖北武汉430074

出　　处：《华中科技大学学报（自然科学版）》2008年第10期52-55,共4页Journal of Huazhong University of Science and Technology(Natural Science Edition)

摘　　要：基于PowerFactorization(PF)仿射重建算法提出了一种最小化重投影误差的线性射影重建算法PFR(PF reprojection).该算法通过线性算法实现了重投影误差的最小化.算法通过添加权值参数将重投影误差非线性的代价函数转化为三线性的代价函数,并基于摄像机运动、三维场景结构以及射影深度之间的加权交错最小二乘法实现了重投影误差的交错最小化.实验结果表明本算法能在不影响输出精度的前提下提升射影重建30%的整体效率,适合作为标准光束法平差算法的引导算法或与光束法平差组成混合算法.On the basis of PowerFactorization（PF）, a linear PowerFactorization reprojection （PFR） method is presented, in which 2D reprojection error was minimized by using projective reconstruction. This algorithm was inspired by Hartley＇s PF method for affine reconstruction. The nonlinear reprojection error was transferred to a trilinear cost function, and a weighted alternated least square method was performed between the structure, motion and projective depth, which can achieve reprojection error minimization. We show experimentally that our method can be used as a bootstrap or composite for the standard bundle adjustment, which speed the overall reconstruction performance up to 30 %.

关 键 词：射影重建 PowerFactorization(PF) 矩阵分解 重投影误差 缺失数据 主元分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]