A Study On Terrain Adaptive Ground Filter For Airborne LiDAR Point Cloud Based On Regularized Thin Plate Spline Interpolation

Airborne LiDAR(Light Detection and Ranging)is an active remote sensing method.It uses laser pulses to measure the information,such as distance to sensors and reflectivity of ground objects.It has all-day and all-weather observation capabilities and can extract valid information about the shape of the earth and accurate threedimensional information on its surface features.LiDAR data can be used to generate a variety of geospatial products,such as Digital Elevation Model(DEM),canopy models,building models,and contour maps.They are widely used in “smart cities”,engineering building surveys,disasters detections and the environmental protection,forestry survey,resource exploration and archeology,thus have broad application prospects.With the technological upgrading and widespread application of civilian UAV,integrating LiDAR equipment on drones for data acquisition has become a more mainstream data acquisition solution,which has increased operating efficiency and significantly reduced costs.The production of DEM products is one of the important applications of LiDAR point cloud data.Point cloud filtering is the process of separating ground point cloud and non-ground point cloud from LiDAR point cloud data.With the development of laser scanning equipment,the point density and volume of point cloud data has increased.It can present more details of terrain and features,laying a foundation for the production of high-precision geospatial data products.In the meantime,it brings new challenges in point cloud filtering methods.Additionally,the efficiency of mass point cloud data processing should be considered.Existing filtering methods are difficult to meet the actual production efficiency needs when processing high-density and largescale point cloud data;problems occurs when we filter point clouds with special terrains or ground objects,such as steep slopes,terrain depressions,slopes containing vegetation and complex buildings,it is prone to cause significant errors.When most filtering methods deal with large-scale point clouds,for point clouds of different terrain environments,it is necessary to repeatedly adjust the filtering parameters and do multiple adjustments according to the quality of the filtering results to achieve filtering results with better usability.This process is time-consuming and complicated.In view of the above problems,this study is carried out from the following aspects:(1)Study the method of interpolating point selection based on two-dimensional region growing and multi-directional scanning of voxel.The first method uses twodimensional region-growth combined with line feature constraints to extract reliable interpolation reference points quickly.The second method uses vertical and horizontal scanning to separate unreliable voxels in the case of a point cloud organized by a voxel grid.And from the remaining voxels,several rules are used to extract reliable ground points as the initial reference points for interpolation.On the basis of implementing the above method,compare the quality of the interpolation points generated by different methods,and whether there is a gap in accuracy.(2)Research on the use of regularized thin plate spline(RTPS)to construct different terrain interpolation surfaces.Use the TPS interpolation filtering method and set the regularization term,and select appropriate point space fluctuation degree indicators from the existing quantitative index of terrain fluctuations through experiments as the terrain constraints of the RTPS,and take flexible measures for the terrain conditions of the survey area and build interpolation classification surface.The research focuses on the use of quantitative terrain fluctuation indicators to generate interpolation surfaces under different terrain and interpolation point distribution conditions.By comparing the quality of the interpolation surfaces,the performance of the quantitative terrain fluctuation indicators is tested.Based on the results of the filtering,the filtering classification accuracy of the point cloud under different terrains and the quality of the generated DEM are compared with this filtering method and the current mainstream mature filtering methods,and the robustness of the method to point clouds with different terrains and different densities.(3)Study the rules and methods of automatic detection and evaluation of low reliability areas based on sparse TIN and elevation kurtosis.The methods are used to detect the areas with low filtering reliability(high probability of filtering errors)in the filtering results,these regions are automatically detected and marked.After that verify whether the two methods are practical and compare the performance of the two methods in the test area.In order to ensure the robustness of the method,experiments were performed using multiple sets of point cloud data.The point cloud data are located in the survey areas of different terrains and have different point cloud densities and points.The data are mainly from the ISPRS 2003 filter test data and point cloud data of a university in Guangzhou.The comparative experimental results of the two sets of data show that:using the method of combining K-nearest neighbors and bottom hat transformation,the point cloud noise points can be removed without significantly increasing the amount of calculation,including isolated noise and clustered low noise points;The methods of area growth and voxel scanning can establish a spatial topology relationship and select a certain number and accurate ground point set from the initial point cloud data,and use it as the initial seed point for the construction of the point cloud classification surface;in comparing a variety of regularized TPS samples.After the strip interpolation method,relative coefficient of variation is used to adapt to the terrain of different fluctuations,and complete the construction of the interpolation classification surface under the condition of different interpolation point reference points.The research process and conclusion provide effective reference and scientific reference for the highprecision automatic processing of massive point clouds.