Study On Airborne Lidar Point Cloud Filtering Algorithm

In recent years,with the constant maturity of airborne laser lidar data acquisition technology,the technology has been widely used in mapping,power,traffic and forestry,and social demand is gradually increasing.LiDAR data processing is an important part of the airbore laser system,accounting for about 80% of the total project time of lidar work,which is basic data processing work of the filtering of point cloud to obtain DEM,and the production and application of the subsequent digital 3D products play a key role, and therefore the filtering of point cloud became an important research topic direction in LiDAR system. Currently filtering LiDAR research focuses on how to improve filtering accuracy and better retain key terrain features with less human intervention by the computer and how to achieve no or fewer input parameters to automate filter in order to achieve the automated filtering. In view of this situation, the main work and innovation of this paper is as follows:1. Study of Mechanism of Point Cloud Filtering AlgorithmThe past few related article describes the prerequisite of realizing filtering algorithm, in the research status of this paper, the current filtering algorithm implementation mechanisms and assumptions conditions are analyzed and presented, while the current filtering algorithm major data processing software and research institutions are summarized, which provide more for the study after filtering algorithm the theoretical basis and resources.2. Study of the comparison of existing point cloud filtering algorithmFilter algorithm is necessary to study the characteristics of the current filtering algorithm, which is the use of improved and innovative algorithms. First,the assessment methods of the current filtering algorithms filtering accuracy are summarized and concluded in this article,on this basis,the selected 12 kinds of current mainstream filtering algorithms are compared and analyzed. For most of the current algorithm does not have universal characteristics, where innovation will be 15 sets of experimental data into three groups: the rough slopes and dense vegetation area; relatively flat urban area;discontinuous surface and rough terrain.By grouping experiments,the characteristics and insufficient of current filtering algorithm can be visually reflected.3. Study of an improved mathematical morphology filtering algorithmMathematical morphology theory is mature,high efficiency,and the Airborne laser lidar filtering algorithm has occupied an important position. In this analysis of the deficiencies of the existing morphological filtering algorithm, filtering algorithm based on the theory of mathematical morphology filtering and TPS model is presented.The features of this method is that mathematical morphology is not used to obtain DEM,but first get an approximate DEM by mathematical morphology, which introduced hole repairing technique due to the hole caucsed he river or lake, this repair technique can significantly reduce the impact on the final point cloud classification. In the end of the original point cloud classification process,where TPS deformation model is introduced, TPS model relative to other interpolation methods have physical properties that reflect abnormal elevation changes with a smooth, continuous, flexible and good features,but the method not need to be regularly distributed point cloud, so this method is more suitable for interpolation of discrete point cloud,where smooth surface will be obtained and will help improve the accuracy of the interpolation. In the classification process, each discrete point will be based on the approximate DEM and use the TPS model interpolation to obtain slope and elevation values shaped seat that point, so the final classification of the original point cloud based on a combination of factors are the ground point and non-point ground, final experiments show that the method has higher precision and better retain terrain features,but also assist the artificial filter to improve filtering accuracy.4. Study of the unsupervised classification of skewness balance method filtering methodSkewness balance method based elevation is a typical unsupervised classification method without input parameters and auto-complete filter. Its assumption is that the ground point give obedience to the normal distribution and non ground point can interfere the distritution, but this skewness coefficient parametric statistical-based methods have some limitations only for relatively flat area,where assume buildings and other non-ground points are above the ground point, so there are ups and downs in the urban terrain and The filtering results will be failure. This paper presents an improved method based on height of skewness balance method in this case,the first terrain will be fit by quadratic surface,then find height between each point and the corresponding approximate elevation of the terrain and use skewness balance method to complete filter based height, the final article is to verify the feasibility of the method through experiments.