面向自由视角的多片三维扫描数据无约束拼接算法

Unconstrained Registration Algorithm for Multiple View 3D Scanned Data

提出一种面向自由视角扫描点云的两步法三维拼接算法。首先在重叠区域采样出两片大致相似的点云,根据空间坐标组合变换原理,使用N边形法进行预拼接,获取预拼接的转换矩阵参数;然后基于初始位姿对采样点云使用改进的ICP算法进行微调精拼,融合传统的迭代最近点距离度量函数的优点,通过分层次设定迭代终止条件和多次重采样获取对应点对,以减少迭代次数,同时保证全局最小收敛。试验结果表明,该三维拼接两步法具有较高的拼接精度和好的操作性,适用于不能进行粘贴标记点和转轴固定的特殊三维视觉测量领域。

A two--stage algorithm for registering 3D scans obtained from multi--view of an object was described. First step, two similar point clouds was selected from overlap regional. According to the principle of space coordinates transform, a new N--edge algorithm was used for pre--registration and acquire the transform matrix parameters of pre--registration. Then, the improved ICP（iterative closest point） algorithm was used to registration accurately for the initial positional sampling point clouds. This second step integrated the merits of traditional iterative closest point distance function, set multi-level thresholds for the termination of iterative and resample the corresponding point level by level. The two--stage technique has advantages in terms of accuracy, efficiency and global mini mum convergence. Experiments show that this two--stage algorithm has the advantage of perfect per- formance and high precision, it is suitable for special 3D measurement registration field without coded points and no mechanical turntable.