视觉SLAM中三维点云配准的Cayley方法

Cayley approach for 3D points registration in visual SLAM

利用三维点对应关系估计摄像机的位姿矩阵是同步定位与地图构建(SLAM)系统的一个基本问题。由于系统的误差累积会导致点云尺度的漂移,三维点对应关系在误差干扰下从欧式变换退化为相似变换,在计算过程中需考虑尺度因子。利用Cayley变换来估计三维对应点之间相似变换的闭式算法,可方便地确定定位与构图过程中摄像机的位姿变化。仿真实验和误差分析验证了基于Cayley变换的配准算法的准确性与鲁棒性。该算法不仅可用于SLAM系统,还适用于计算机视觉中的三维重建与三维拼接问题。

An essential problem in SLAM system is to compute the 3D rigid body transformation of cameras, which aligns two sets of points when correspondences are known. The errors accumulated during the locating and mapping process result in the drift of scale for the pairs of points. The Euclidean transformation will be degraded into the similarity transformation due to the drift. A closed-form solution using Cayley transform is presented to estimate the similarity transformation between 3D corresponding points. The accuracy and robustness of the proposed Cayley approach are verified by simulation and error analysis. Apart from SLAM system, this approach is also suitable for 3D reconstruction and registration in computer vision.