Msplit Estimation with Local or Global Robustness Against Outliers—Applications and Limitations in LiDAR Data Processing

Light Detection and Ranging (LiDAR) systems become more prevalent in remote sensing for modeling buildings, engineering structures, or their deformations and displacements. Processing data from such systems, usually point clouds, can be performed using different methods, including Msplit estimation. The method in question is relatively novel but it has several variants. From a practical point of view, the variants that are globally or locally robust against outliers seem very promising. The paper addresses robustness and the problem of different types of outliers that might disturb LiDAR point cloud processing by Msplit estimation. The basic variants, the squared and the absolute Msplit estimations, are often sensitive to global outliers and cannot always deal with local outliers. The comparative analyses show that the modifications of the basic Msplit estimation variants complement each other. Hence, one can always find an Msplit estimation variant that is appropriate for processing LiDAR data disturbed by different types or share of outliers. The paper points out such variants and their application range. It also gives clues on using the methods in question in practice.