Research On Dense Matching Point Cloud Filtering Method For Oblique Image

In recent years,with the rapid development of computer vision technology and parallel computing CPU,inclined photogrammetry in photogrammetry technology has been perfected and improved,becoming one of the current research hotspots.Oblique photogrammetry breaks through the bottleneck of traditional photogrammetry,and can comprehensively obtain three-dimensional surface data information.The surface three-dimensional data information is composed of matching point clouds,which contain not only ground points indicating terrain fluctuations,but also non-ground points such as buildings and vegetation.To extract the ground point production digital elevation model from the matching point cloud,a point cloud filtering algorithm is needed to distinguish ground points from non-ground points.Most of the existing point cloud filtering algorithms are designed for Li DAR point clouds.However,because the oblique image matching point cloud acquisition mode is different from the Li DAR point cloud,having the following characteristics: more number of point clouds,greater density of point cloud,discrete form distribution,and both a small amount of large error in the point cloud and a large number of noise points with minor errors.These characteristics bring great difficulties to the dense matching point cloud filtering of oblique image,which makes the LIDAR point cloud filtering algorithm not suitable for the processing of the matching point cloud of oblique image.Especially when the point cloud filtering is carried out in a variety of mixed terrain scenes,the classified ground points are more prone to be misclassified and omitted,which affects the efficiency and accuracy of the point cloud filtering.Therefore,it is of great research significance to study the high automation and high precision filtering algorithm for the dense matching point cloud of oblique image.Using the classical cloth simulation filtering method,the article aims to further improve the effect of cloud filtering of oblique image in various mixed terrain scenes.The main research contents of this article are as follows:(1)Research the matching cloud acquisition method of oblique image,focus on the key technologies of aerial triangulation and dense matching in the process of oblique image matching point cloud acquisition,and summarize the characteristics of oblique image dense matching point cloud by combining the characteristics of oblique image and dense matching principle.(2)Research the filtering idea and method of classical filtering algorithm,carry out the filtering experiments for LIDAR point cloud and oblique image matching point cloud using the relevant data set,and compare its adaptability in oblique image point cloud filtering.(3)Improve the cloth simulation filtering algorithm.Aiming at the problem of error division and omission of ground points after point cloud filtering classification in various terrain mixed scenes,based on the method of classical cloth simulation filtering,the cloth simulation filtering method based on virtual grid classification is proposed.The method firstly induce a virtual grid to distribute the zone,and the statistical analysis filtering method is used to eliminate the mismatching outlier in the grid.At the same time,in order to solve the problem of error caused by not close fitting of adjacent grid cloth filtering,a buffer is added on the basis of virtual grid;Secondly,classify the virtual grid by slope,and according to the cloth parameters of cloth simulation filtering,it can be divided into flat terrain,gentle slope terrain and steep slope terrain;Then,the point cloud in the virtual grid is filtered by setting cloth parameters and height threshold according to its own terrain;Finally,the ground point cloud after the virtual grid filtering is fused to get the ground points of the entire area.The experiment show that the cloth simulated by this method is closer to the actual ground of mixed terrain,solving the problem of error division and omission of ground points due to the cloth internal force is not suitable for the actual terrain,and the efficiency and accuracy of the filtering are further improved to a certain extent.The method proposed in this thesis is suitable for separating ground and non-ground points in dense matching point clouds of oblique image in a variety of mixed terrain scenes.