基于图优化的增量式单目SFM三维重建方法

doi:10.1004-7530/2019-36-6-37

摘要/Abstract

摘要：

相机位姿估计作为SFM的关键环节,其估计精度直接影响三维重建结果,而在增量式SFM的迭代过程中,经常出现由于相机位姿估计误差累积而在最终重建结果中产生漂移问题,文章将图优化理论引入增量式SFM相机位姿估计过程中,以重投影误差平方和为代价函数构造图优化模型,对估算的相机姿态和重建的三维点云进行优化。实验表明本文方法可达到SFM重建要求,且漂移问题得到了明显改善,重建结果具有良好的视觉效果。

关键词: 增量式SFM, 图优化, 三维重建

Abstract:

Camera pose estimation is a key part of SFM（Structure From Motion）, and its accuracy direct affects the results of 3D reconstruction. However, in the iterative process of incremental SFM, there often occurs drift due to camera position and pose estimation error and drift in the final reconstruction results. This paper introduces graph-based optimization theory into the pose estimation process of incremental SFM camera. The graph optimization model is constructed by using the sum of the re-projection errors as the cost function to optimize the estimated camera pose and the reconstructed 3D point cloud. Experiments show that the proposed method can meet the requirements of SFM reconstruction, and the reconstruction results have good visual effects.

Key words: incremental SFM, graph-based optimization, 3D reconstruction

中图分类号:

段建伟. 基于图优化的增量式单目SFM三维重建方法[J]. 江苏科技信息, 2019, 36(6): 37-40.

Jianwei Duan. Incremental monocular SFM 3D reconstruction method based on graph optimization[J]. Jiangsu Science & Technology Information, 2019, 36(6): 37-40.

图/表 5

参考文献 15

[1]	曹明伟, 李书杰, 贾伟 , 等. 运动推断结构技术中的特征跟踪方法综述[J].计算机学报, 2018(11):1-32.
[2]	曹明伟, 李书杰, 贾伟 , 等. 面向大规模SFM的快速鲁棒特征跟踪算法[J].仪器仪表学报, 2017(1):252-260.
[3]	薛武, 张永生, 赵玲 , 等. 增量式SFM与POS辅助光束法平差精度比较[J].测绘学报, 2017(2):198-207.
[4]	WU C C . Towards linear-time incremental structure from motion [C].Seattle:Proceedings of the IEEE International Conference on 3D Vision-3DV, 2013: 127-134.
[5]	WU C C . VisualSFM:A Visual Structure from Motion System[J/OL].2011:1-5[ 2019- 02- 16]. .
[6]	GALLUP D, GOLOVINSKIY A, LIN M H , et al. Combining narrow-baseline and wide-baseline stere for three-dimensional modeling[P].United States Patent 8463024:2013.
[7]	方磊 . 基于特征的图像序列三维场景重建技术研究[D]. 武汉:华中科技大学, 2007.
[8]	陈辉, 马世伟, ANDREAS N . 基于激光扫描和SFM的非同步点云三维重构方法[J].仪器仪表学报, 2016(5):1148-1157.
[9]	史利民, 郭复胜, 胡占义 . 利用空间几何信息的改进PMVS算法[J].自动化学报, 2011(5):560-568.
[10]	KLEIN G, MURRAY D . Parallel tracking and mapping for small AR workspaces [C].Nara:IEEE and ACM International Symposium on Mixed and Augmented Reality, 2007 : 225-234.
[11]	COLBERT M, BOUGUET J Y, BEIS J , et al. Building indoor multi-panorama experiences at scale[J].ACM Siggraph, 2012(2):101-102.
[12]	ARRIGONI F, ROSSI B, MALAPELLE F , et al. Robust global motion estimation with matrix completion[J].International Archives of the Photogrammety Remote Sensing & Spatial Information sciences, 2014(5):63-70.
[13]	SONG S, CHANDRAKER M . Robust scale estimation in real-time monocular SFM for autonomous driving [C].Culumbus:Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2014: 1566-1573.
[14]	鲁晨曦 . 基于图像的全局SFM三维模型重建方法研究与实现[D]. 成都:电子科技大学, 2017.
[15]	高翔, 张涛 . 视觉SLAM十四讲:第十讲[M]. 北京: 电子工业出版社, 2017.

影像编号	图优化SFM	VisualSFM
0	0.301	0.946
1	0.373	0.792
2	0.556	0.795
3	0.584	0.694
4	0.452	0.35
5	0.449	0.6
6	0.463	0.421
7	0.395	0.35
8	0.385	0.946
9	0.471	0.967
10	0.500	0.993
11	0.360	0.936
12	0.434	0.878
13	0.327	0.934
14	0.265	0.92

《江苏科技信息》杂志社
Jiangsu Science & Technology Information Magazine Agency