具有LTS稳健初值的点云平面拟合算法

Plane Fitting of Point Clouds with Robust LTS Initial Value

摘要:

针对粗差含量较高的点云平面拟合，提出具有最小截断二乘（LTS）稳健初值的拟合算法。首先采用随机抽样的方法计算平面方程系数的LTS估值，选择稳健的绝对偏差中位数作为初始单位权中误差，然后再进行选权迭代。这样既保证了选权迭代法较高的估计效率，又可以保持LTS方法的稳健性。实验结果表明，具有LTS稳健初值的选权迭代法对于粗差含量较高的点云数据仍然稳健，可以得到准确的平面方程。

Abstract:

To fit a plane with point clouds containing massive outliers, a fitting method with robust initial value by least trimmed squares (LTS) was presented. It firstly used LTS to ascertain initial plane coefficients by random sampling. Meanwhile, the median of all absolute deviations was adopted as initial mean square error of unit weight. Then it carried out selecting weight iteration. This method kept the high efficiency of selecting weight iteration as well as maintained the robustness of LTS. The test results show that selecting weight iteration with LTS initial value can keep robust to deal with point clouds including massive outliers. Finally, the accurate plane can be fitted.