Abstract
Robust plane fitting is very important and has been well-studied in mathematics, computer vision, CAGD, and many other applications including road inspection from scanned point sets. Previous studies found that principal component analysis gives us better results compared with naive least-squares fitting, but both of them suffer from outliers. Therefore, random sampling and median strategies have been successfully employed to eliminate such outliers. In this research, we propose a robust plane fitting method based on weighted median covariation matrices with stochastic gradient descent.
