Research On Methodology Of Filtering And Building Extraction Based On Airborne Lidar Data

The Airborne light detecting and ranging(Lidar)technology is a new type of surveying and mapping technology that can quickly acquire topographic information through airborne radar equipment.This technology has obvious advantages,including short work cycle,high data accuracy,strong environmental adaptability.It is therefore widely used in forestry detection,terrain mapping,digital cities,railway power lines planning and other fields.In recent years,the research using airborne laser radar data has focused on the key issue of how to separate ground and ground object information quickly and accurately.Therefore,this dissertation takes the airborne laser radar data as the research object,and deeply studies the key issues such as the filtering classification of airborne laser radar data and the extraction of building point clouds.The main research contents of this article are as follows:1)It systematically introduces the composition and working principle of the airborne Lidar system,and the concept and characteristics of LiDAR point cloud data.In addition,a comprehensive overview of the advantages and disadvantages of commonly used filtering algorithms and related research on feature extraction and classification provides a theoretical basis for this study.2)Aiming at the existence of high-low coarse-difference points and isolated noise points in the original point cloud data,this paper proposes an improved algorithm based on local K-nearest neighbor denoising,which fully considers the elevation difference between the measured point and its neighbors,and the geometric distance from the nearest neighbors.This method can quickly and effectively eliminates the high and low level gross and differential noise points and isolated noise points in the point cloud data,and also has a certain effect on the clustered high and low positions.3)This paper proposes a point cloud filtering method of adaptive strip strategy.By dividing the 3D point cloud data into x and y direction strips,the calculation is reduced from 3D to 2D,and the data strips are further sliced in consideration of the terrain conditions.Generate grid strips and perform polynomial fitting on each stripe grid.Then calculate the residuals of the real data and the fitted curve,and filter the point cloud according to the adaptive threshold of the terrain to obtain the classification result.Finally,the final filtering results are obtained by combining the classification results of the data points in the x and y directions.Through a series of experiments,this method can effectively filter point cloud data and has better filtering performance than existing filtering algorithms.4)According to the traditional method of determining the vegetation point using the elevation difference of the first and last elevation differences between the measured point and other first return points in its neighborhood to determine whether the multiple return points are vegetation point.Experiments show that this method can effectively detect the vegetation points and avoid the misclassification of the building edge return points into vegetation points by calculating the standard deviation of the by calculating the standard deviation of the elevation difference of the return points in the neighborhood of the measured first return point.5)A method of building edge extraction based on local multi-feature classification is proposed.Firstly,R radius neighborhoods are constructed for each non-ground point,multi-feature information is extracted for each neighborhood set,and then multi-feature information is used.Feature classification,clustering neighborhoods based on neighborhood connectivity,and finally extracting building edge point sets using an improved scanline algorithm.Among them,in this method,the classification coefficient of the laser spot is constructed to determine the feature classification,and the laser spot is marked as building point,vegetation point and other points by calculating the classification coefficient of each laser spot.After laser spot classification,each cluster then separates the building point cloud and the vegetation point cloud according to the classification of the laser spot.Through experimental verification,the local multi-feature-based classification method can effectively segment and extract the building point clouds of urban laser radar data,and has a good classification effect on vegetation points.In addition,the improved point cloud scanning line algorithm can accurately extract Building edge.