Aimed at tackling the fast solver problem for the large-scale and nearly pathological close-range block sparse bundle adjustment normal equation,the authors propose a solution method based on the preconditioned conjugate gradient(PCG)sparse algorithm. Firstly,the normal equation coefficient matrix corresponding to diagonal matrix square root is selected as the preconditioning matrix by changing the coordinate base of the parameter vector to be estimated, which can improve the behavior of the normal equation coefficient matrix so as to achieve the purpose of improving the convergence rate of the conjugate gradient method. Then, through the application of the sparse matrix, the efficiency of storage can be improved and the adjustment of normal equation coefficient matrix can be achieved. Experiment results show that the method proposed in this paper has the advantage that any scalar change in variables has no effect on the range of convergence of the iterative technique,and hence it has not only high accuracy of calculation but also faster speed.
[1] Lourakis M,Argyros A.The Design and Implementation of a Generic Sparse Bundle Adjustment Software Package Based on the Levenberg Marquardt Algorithm[R].ICS/FORTH Technical Report,No340,2004.[2] Cornou S,Dhome M,Sayd P,et al.Bundle Adjustment:A Fast Method with Weak Initialisation[G].Cardiff:BMVC,2002:223-232.[3] Bartoli A.A unified framework for quasi-linear bundle adjustment[C]//Proceedings of the 16th International Conference on Pattern Recognition.Quebec City,Quebec,Canada:IEEE,2002,2:560-563.[4] 冯其强,李广云,李宗春.基于点松弛法的自检校光束法平差快速计算[J].测绘科学技术学报,2008,25(4):300-302. Feng Q Q,Li G Y,Li Z C,et al.Speedy calculation of self calibration bundle adjustment in digital industrial photogrammetry[J].Journal of Geomatics Science and Technology,2008,25(4):300-302.[5] Olsson C,Kahl F,Oskarsson M.Optimal estimation of perspective camera pose[J].International Conference on Pattern Recongnition,2006,2:5-8.[6] Kahl F,Henrion D.Globally optimal estimates for geometric reconstruction problems[J].International Journal of Computer Vision,2007,74(1):3-15.[7] 朱肇光.摄影测量学[M].2版.北京:测绘出版社,1995. Zhu Z G.Photogrammetry[M].2nd ed.Beijing:Surveying and Mapping Press,1995.[8] 徐振亮.轴角描述的车载序列街景影像空中三角测量与三维重建方法研究[D].武汉:武汉大学,2014. Xu Z L.Research on Aerial Triangulation Angle/Axis Representation and 3D Reconstruction for Vehicle-borne Street-level Image Sequence[D].Wuhan:Wuhan University,2014.[9] 吴建平,王正华,李晓梅.稀疏线性方程组的高效求解与并行计算[M].长沙:湖南科学技术出版社,2004. Wu J P,Wang Z H,Li X M.Efficient Solving Sparse Linear Equations with Parallel Computing[M].Changsha:Hunan Science and Technology Press,2004.[10] Davis T.Direct Methods for Sparse Linear Systems[M].Philadelphia:SIAM,2006.[11] Dellaert F,Kaess M.Square root SAM:Simultaneous localization and mapping via square root information smoothing[J].International Journal of Robotics Research,2006,25(12):1181-1204.