Research On Adaptive Filtering Method Of Unmanned Aerial Vehicle LiDAR Points On Cloud Mountain

Unmanned Aerial Vehicle-Light Detecting and Ranging(UAV-LiDAR)is a new technology that uses UAV as a carrier and is equipped with hardware such as a 3D laser scanner to actively obtain high-precision 3D spatial information.It has been widely used in topographic surveying and mapping,Land and Resources Survey and other fields.At present,the development of UAV-borne LiDAR hardware equipment has been relatively complete,but the algorithm for point cloud data post-processing is still relatively lagging.One of its most important data processing procedures is point cloud filtering.Although many scholars have studied it,there are still some problems in these filtering methods,such as single selection feature,poor adaptability and poor applicability to mountains with high vegetation coverage,which brings challenges to accurate filtering.Based on this,this paper focuses on the UAV-borne LIDAR point cloud filtering method,and proposes an adaptive double filtering method for point cloud in densely planted areas.The main research contents are summarized as follows:(1)In view of the noise points and abnormal points in the original LiDAR point cloud,some relatively obvious high-level and low-level noise points are eliminated through visual inspection,and the intensity information is used for frequency statistical analysis to eliminate the intensity abnormal values.(2)Aiming at the problems of single features and poor adaptability in the current filtering methods,this paper firstly conducts multi-echo separation for the point cloud data,and through the visual discriminant analysis of echo separation and surface objects,the single echo and last echo of point cloud are extracted respectively.Then,according to the difference between single echo and multiple echo in the dense vegetation covered area,different attribute features were selected to filter the single echo and multiple echo of point cloud simultaneously based on skewness balance and OTSU method.Finally,automatic threshold setting is realized to complete adaptive coarse filtering of point cloud.(3)In order to solve the problem of uneven point cloud data distribution after rough filtering,the idea of block is introduced,and the size of grid is calculated by using the density of local point cloud of block,so as to realize adaptive extraction of seed points of terrain filtering.This filtering method improves the reliability of ground seed points and is more conducive to data filtering.(4)For topographic relief and other problems,The Thin plate spline interpolation function(TPS)considering local details was selected to fit the terrain surface,and the regularization coefficient was introduced to achieve more accurate terrain fitting.Firstly,the function coefficient of the improved fitting surface is solved by obtaining the ground seed points.Then,the bending energy minimization property of the thin plate spline interpolation function is used to substitute the waiting point into the function to calculate the fitting elevation,and according to a certain constraint criteria to determine whether the waiting point is a ground point,and finally achieve the precise filtering of point cloud data.Finally,the experimental results processed by the method in this paper are analyzed both qualitatively and quantitatively.The experimental results show that the UAV point cloud filtering method in this paper has achieved a good filtering effect,which lays a foundation for the subsequent generation of high-precision DEM.