Filtering Method Of Airborne Lidar Data Based On Skewness Balancing Of Elevation Information And Terrain Structure Feature

Airborne LiDAR is a surveying system combining laser ranging (LDM), global positioning system (GPS) and inertial navigation system (INS) together, which is used to obtain three-dimensional data on earth surface. We can do some process to obtain DEM, DOM. DSM, and other bountiful data products by using laser point cloud data acquired directly and the digital image correspondingly. The point cloud data filtering refers to the separation of the ground point set from point cloud, which is an important part of data post-processing and the basis of related data products. With the development of the system hardware and people’s requirements for more precision and higher accuracy constantly, filtering algorithm is facing mass data, various terrain, complex topographic features and other challenges. There appeared a lot of ideas, improved methods and research results after the ISPRS’s test analysis for classic filtering algorithms in2003, but the difficulty to determine threshold and the poor stability in complex test area are the main problems existing in most of filtering algorithms. Therefore, the study for really accurate, efficient and adaptive algorithm has certain value and significance. Faced with above problems, it has improved the Bartels’ skewness-balancing algorithm through summarizing the filtering algorithm in existence, and conducted an experimental analysis using the reference data samples providing by ISPRS. This paper carries out following work:(1) The skewness-balancing filtering algorithm based on elevation information is studied and the relevant improvements of this algorithm by scholars at home and abroad are summarized. Aiming at the existing problems, such as the influence of the reliability of the sample statistics to points’ number, termination conditions of the balance is difficult to meet, the applicability of the iterative algorithm and region and lack of terrain detail processing, an improved algorithm is proposed by removing high point from high to low is switched to pulling them form low to high, changing the terminal condition from ’sk<0’ to ’sk<e’(e is a very small number), which enhance the reliability, rate of convergence and adaptation in hillv area.(2) The design idea and relevant improvement of classic algorithms are summarized. It focuses on the gradually encryption algorithm based on TIN and improves the conventional algorithm by terrain feature points extraction, quadtree grid partition, seed point selection under the restriction of terrain feature points, adding some auxiliary border points, priority to encrypt the grid with characteristics and sorting of the point to be determined to make up for the inadequacy that low terrain is difficult to eliminate and its easy to destroy the terrain. It enhances the accuracy of judgment for undetermined points, the filtering effect of low terrain and is able to protect the terrain features in hilly area.(3) Two typical point cloud data represents moderate and hilly terrain are chosen respectively, then conducting Filtering experiments using the improved algorithm, finally comparing with the filtering result obtained by original algorithms to evaluate the quality of the improved algorithm from the perspectives of two types of errorIt shows that the proposed filter algorithm is feasible. it combines the advantages of the two algorithms together. Specifically, the moderate point cloud will obtain a better effect using the gradually encryption algorithm based on TIN after being split into tiles by skewness-balancing. However, it spite hilly point cloud from bottom to top to ensure that points are encrypted into TIN from low to high, and it restrict initial TIN by terrain feature points extracted through the analysis of the contour to make the process does not destroy the terrain structure with stronger threshold and able to filter out the low terrain in each elevation range. The two types of error analysis shows that this algorithm have a good control of two types of error in two types of point cloud, and it can reduce the type I error on the premise of stabilizing the type Ⅱ error.