基于移动最小二乘的数据拼接ICP算法

ICP Algorithm of Data Registration Based on Moving Least-squares Methods

ICP算法的关键是确立正确的对应点集.针对此问题,本文提出一种改进的ICP算法.首先对2片点云的重叠区域进行基于移动最小二乘(Moving Least-Squares,MLS)曲面拟合得到2片插补出的点云数据,并确立初始对应点,然后结合刚性、曲率等约束去除错误对应点对.实验结果表明:改进ICP算法可以有效地提高确立对应点的正确率,从而极大地提高数据拼接的精度.

It is crucial for the ICP algorithm to establish the right corresponding points set.To solve this problem,an improved algorithm of ICP was introduced in this paper.First,the point clouds of dual interpolation were required for partially overlapping point clouds and the original correspondences were established based on moving least-squares methods to fit surface.Then,rigid constraints,together with surface curvature information,were employed to remove unreliable matches.As the experimental results demonstrated,for data registration,the ICP variant presented in this thesis could efficiently reject false correspondences to obtain high quality refined correspondences,thus greatly improved the accuracy of fine registration for point clouds.