Research On DEM Generation From Point Cloud Data Based On Combined Filtering Algorith

With the advancement of information technology,the digital expression of geospatial data has been further developed,and the frequency of digital geographic information updates has become faster and more frequent.As an important part of spatial data infrastructure and digital geospatial data framework,digital elevation model is the key supporting data in the fields of geoscience research,geological disaster analysis and national defense security.It has great strategic value and economic value.How to generate high-precision digital elevation model quickly,conveniently and widely is one of the important research contents in the field of surveying and mapping.Compared with the traditional three-dimensional acquisition method,airborne lidar,as an emerging active remote sensing observation technology,can quickly obtain highprecision and high-density three-dimensional spatial information in large-area measurement areas,and has become the preferred technical means to obtain digital elevation models.The point cloud data obtained by airborne lidar technology has the characteristics of large amount of data,redundant information,and inability to directly obtain surface topology information.Filtering point cloud data and generating DEM is the basis and premise of point cloud data application,and it is also one of the current research hotspots.Most of the existing filtering algorithm research has problems such as the need for manual interaction to set parameters,low degree of automation,and insufficient filtering accuracy.Because the filtered point cloud data eliminates the nonground points,there is a data missing hole,and it is necessary to select the appropriate interpolation method to generate a high-precision digital elevation model.In view of the above problems,this thesis makes a systematic and in-depth analysis and research on the digital elevation model generation process based on LIDAR point cloud data.Firstly,the acquired point cloud data is denoised,and then the improved adaptive threshold filtering algorithm and the applicability of digital elevation model interpolation algorithm are focused on.The main research work and conclusions include the following four aspects:(1)This thesis systematically expounds the composition and principle of airborne lidar system,summarizes and analyzes the current research status of point cloud data filtering algorithm and digital elevation model interpolation algorithm.On the one hand,by introducing the scanning principle and method of airborne lidar system,the composition and data characteristics of point cloud data are understood,which provides a theoretical basis for subsequent data processing.On the other hand,based on the research status of point cloud filtering algorithm,the characteristics and applicability of existing algorithms are compared and analyzed,which provides an improved idea for the filtering algorithm in this thesis.(2)In the process of point cloud denoising,an improved statistical filtering denoising method based on Kd-tree is constructed to solve the problems of large memory and long time consumption caused by the large amount and disorder of point cloud data.The Kd-tree spatial index method is used to effectively manage the sorted point cloud data,which reduces the space complexity of the subsequent denoising algorithm.The improved method eliminates the outlier data that are not in the standard interval by calculating the average distance between each point and the adjacent point.The experimental results show that the method filters out a total of 1304 noise points,the denoising effect is obvious,and the terrain and ground features are well preserved.(3)Aiming at the problem of low accuracy and poor adaptability of most filtering algorithms in complex mixed terrain,an adaptive threshold filtering algorithm considering terrain is proposed based on cloth simulation filtering algorithm and surface fitting algorithm.Firstly,the point cloud data is segmented into grids.For the segmented point cloud data,the least square fitting method is used to construct the trend surface,retain the basic undulation of the terrain,and reduce the number of iterations when the cloth falls.Then,according to the properties of the segmented point cloud data,the adaptive parameters are set.Finally,by calculating the elevation difference between the true elevation and the fitted trend surface as the judgment condition of the distance threshold,the accuracy of the filtering results is further improved.Five regions with different terrain features are selected and compared qualitatively and quantitatively with the existing three filtering algorithms.The algorithm in this thesis slows down the phenomenon of excessive filtering,and can accurately distinguish between ground points and non-ground points.It has high accuracy and strong adaptability with less manual intervention.(4)There are many factors affecting the accuracy of DEM,among which interpolation method is the main factor.According to the characteristics of DEM interpolation method,the applicability of interpolation method in different terrain and data distribution is studied.The filtered point cloud data are interpolated by inverse distance weighting method,ordinary kriging method and point-by-point interpolation method,and the interpolation results are compared and analyzed qualitatively and quantitatively.The results show that the three methods can meet the first-level accuracy index in flat and hilly terrain.In mountainous terrain,ordinary Kriging method can only meet the second-order accuracy index.The three methods can meet the production requirements in general.The characteristics of DEM generated by different terrain details are different.In practical applications,it is necessary to select the appropriate interpolation method to generate DEM model according to the requirements.Finally,the generated DEM is applied to practical engineering to realize the segmentation of individual trees and the extraction of their heights.