大场景三维重建中高精度相机位姿估计  

High-precision camera pose estimation in a large-scene 3D reconstruction

作　　者：龚志杰 陈崇成[1] GONG Zhijie;CHEN Chongcheng(Key Laboratory of Spatial Data Mining and Information Sharing of Ministry of Education,National Engineering Research Centre of Geospatial Information Technology,Fuzhou University,Fuzhou 350108,China)

机构地区：[1]福州大学地理空间信息技术国家地方联合工程研究中心,福州350108

出　　处：《智能计算机与应用》2022年第9期154-159,共6页Intelligent Computer and Applications

基　　金：国家重点研发计划项目(2017YFB0504202)。

摘　　要：针对大场景三维重建中的高精度相机位姿估计问题,提出了一种基于增量式运动结构估计的相机位姿估计算法。算法采用多种约束条件,去除图像特征误匹配,为增量式运动结构估计提供良好的匹配点对集合;引入EPnP方法解决3D-2D问题,提高算法鲁棒性与稳定性;利用光束法平差进行分块优化与全局优化,防止算法陷入局部最优状况。实验结果表明,该算法计算得到的相机位姿参数具有较高的精度。In the face of the problem of high-precision camera pose estimation in 3 D reconstruction of large scenes, the camera pose estimation algorithm based on incremental motion structure estimation is proposed. A variety of constraints are used to remove the false matching of image features, which provides the good matching point pair set for incremental motion structure estimation. EPnP method is introduced to solve the 3 D-2 D problem, and the robustness and stability of the algorithm are improved. Thereafter, the bundle adjustment is used in block optimization and global optimization to prevent the algorithm from falling into local optimum. Experimental results show that the camera pose parameters calculated by this algorithm have high accuracy.

关 键 词：特征匹配 初始化 对极几何 EPnP 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]