摘要
本文提出一种基于图模型的视觉SLAM系统误差动态补偿方法,可以得到更加精确的位姿估计与三维重建结果。该方法首先利用改进生成树遍历得到包含完整场景的最小关键帧子集;然后通过四叉树均衡化算法进行三维地图点选取,并采用基于双窗口约束的平差策略限制计算量;最后按照自检校光束法平差原理构建误差方程进行求解。实验结果表明,该方法能够在保持计算效率的前提下,准确、稳健地补偿系统误差,提升位姿估计精度与重建模型的内符合一致性。
In this paper,a graph model based system error dynamic compensation method of visual SLAM is proposed,which can obtain more accurate pose estimation and 3D reconstruction results.Firstly,an improved spanning tree traversal is used to acquire the minimum keyframe subset containing the complete scene.Secondly,a quadtree-based uniform distribution algorithm is implemented to select 3D map points from each keyframe,after which,a double-window-constraint adjustment strategy is utilized to limit the computational burden.Finally,error equations are constructed and solved according to the principle of self-calibration bundle adjustment.Experimental results show that the proposed method can compensate system error precisely and robustly while maintaining computational efficiency,which helps to improve the accuracy of pose estimation as well as consistency of the reconstructed model.
作者
张一
姜挺
江刚武
谭振宇
袁铭阳
ZHANG Yi;JIANG Ting;JIANG Gangwu;TAN Zhenyu;YUAN Mingyang(Beijing Institute of Remote Sensing Information,Beijing 100192,China;Information Engineering University,Zhengzhou 450001,China)
出处
《测绘科学技术学报》
北大核心
2019年第4期388-393,共6页
Journal of Geomatics Science and Technology
基金
国家自然科学基金项目(41501482
41471387)
关键词
视觉同时定位与地图构建
相机内参数
自检校光束法平差
系统误差补偿
位姿估计
visual simultaneous localization and mapping
camera intrinsic parameters
self-calibration bundle adjustment
system error compensation
pose estimation
