Research On Filtering Algorithm Of LiDAR Point Cloud Elevation Data

LiDAR(Light Detection and Ranging)is remote sensing technology obtaining accurate point cloud information in a large area.It is widely used in terrain surveying and mapping,height estimation,power line monitoring,and many other aspects.However,as the airborne Li DAR point cloud data are used to derive ground topography,the pointcloud data are mixed with measurements from terrain and nonground targets(e.g.,tree canopies).Dividing the points into the ground and nonground ones is needed before creating the topography.The dividing process is called filtering.Although many filtering algorithms are studied,they generally do not perform well in complex terrain with steep slopes and discontinuity features.Therefore,this study aims to improve the existing filtering algorithm in the complex landscape and produce accurate DEM.The main work includes two parts.(1)In improving the classic Progressive Triangulated Irregular Network Densification(PTD)algorithm in steep or discontinuous complex terrain,an improved PTD algorithm based on multi-level pseudo grids and morphological method is studied.It uses the multi-level pseudo grids to obtain more ground seed points and morphological opening operations to re-screen the ground points after iterative steps.In assessing the improved algorithm,four benchmark datasets provided by the ISPRS Commission ?,Working Group ?,were used.The algorithm retains terrain features well,having smaller type ? and total errors in all the datasets than those from the classic algorithm.The maximum values of the two kinds of errors produced by the improved algorithm are only3.05% and 4.22%,respectively,which is less than one-fifth of the results of the classic algorithm.(2)To solve the shortcomings of the classic PTD algorithm's insufficient processing capacity for complex terrain and sensitivity to global uniform parameters,an improved PTD algorithm based on segmentation and terrain adaptive parameters is proposed.The threshold parameters are calculated according to the terrain,and the region-growing technology is utilized to segment the original point cloud to obtain more sufficient ground seed points.In the assessment using the standard test datasets provided by the ISPRS,the algorithm also retains the terrain feature well.Both type ? and total errors were below 7%.The type ? error is less than 20%.Subsequent examinations show that the algorithm is robust.