改进的鲁棒迭代最小二乘平面拟合算法

An Improved Robust Method for Iterating Least-Squares Plane Fitting

摘要:

针对迭代特征值最小二乘法不具备鲁棒性,提出一种改进的统计分析方法,用于含有大量异常点的点云的平面拟合.首先由移动最小二乘法拟合抽样点的近邻域平面,采用最小平方中位数法选择拟合模型,将该模型作为初始模型调用迭代特征值最小二乘法对点集拟合,通过逐渐剔除异常点,不断精炼模型,最终得到较精确的平面模型.此算法克服了一般向后剔除方法的缺点,具有了鲁棒性,且不失原方法的精确性,同时提高了迭代收敛速度.

Abstract:

The iterating eigenvalue least-squares is not robust,so an improved statistic analysis method is introduced for fitting a plane to point clouds containing a great amount of outliers.The method is robust iterating least-squares (RILS).Firstly,some plane models are fitted to the local neighborhoods around sample points by moving least-squares (MLS),and then a good model is selected from those models by least median of squares (LMS)and refined to appropriate the whole point set through eliminating those outliers by iterating eigenvalue least-squares.Different from other backward ways,this method is robust,which retains the accuracy of the original method,and furthermore accelerates the convergence of iteration.