A New LiDAR Filtering Method Based on Multi-Layer Two-Class Segmentation

There are two kinds of filtering methods for airborne LiDAR, one is discrete point cloud based; the other is two-dimension depth images based; the basic idea of which is applying image processing approaches into image segmentation. That is, firstly conversing discrete point cloud data into the two-dimension depth image, and then using various methods of image processing and segmenting out the ground points. In this paper, a new LiDAR filtering method based on multi-layer two-class segmentation is put forward. The main idea is that generating three-dimension digital surface model (DSM) from raw LiDAR point cloud data and obtaining the main feature-type number N firstly. Secondly, using the global threshold segmentation method for two-class incomplete classification, namely high-lighting the main feature category, and using the average pixel value of the outlying regions of the main feature to replace the main features region and generating a new two-dimension depth image. Finally repeating the procedures above for the times of the main feature-type number N and acquiring DEM with most of features filtered. This approach is easy-doing and very useful for some large-scale engineering construction, such as the highway designation. Take the plain areas for an example, some beneficial suggestions are given.