Discrimination Of Error Points In Aerial Stereo Image Dense Matching Point Cloud And Improvement Of The Dense Matching Method

In recent years, as the infrastructure of city informatization construction, digital city received more and more attention. Three dimensional city models (3DCM) are the core of the digital city, but the three-dimensional modeling of the city is a time-consuming and laborious work. Using high-resolution aerial stereo images to build 3D models of cities through automatic matching of stereoscopic images is feasible technique approach, it's technology method has been deeply researched, and also developed a variety of software products, such as Smart3D, Pix4D, etc. However,due to the presence of texture in the image is not rich or texture repetition,occlusion,parallax discontinuity and other factors, there inevitably exist mismatching, but these software didn't judge the error points. So we need to identify and eliminate the error points, and then improve the image dense matching method on the basis of analyzing the distribution characteristics of the error points.The method of judging the error points in the aerial stereo image dense matching point cloud is deeply studied in this paper, and then improved the method of dense matching. The main research contents and achievements are as follows:(1) Theory and method of aerial stereo image matching. Describe the process of image matching process: matching measure, matching primitive and matching strategy, and then discussed the classical methods in aerial stereo image matching like SIFT feature matching and least squares image matching.(2) Stereo image dense matching and point cloud generation. Using per-pixel dense matching of the stereo images on the basis of the SIFT feature matching. First,determine the image search range according to the epipolar constraint conditions; and then using the correlation coefficient method to calculate the similarity measure;choosing the point whose similarity measure is optimal as the temporary matching point; performing a left and right consistency check on the temporary match point to choose final match point; finally generate 3D point cloud data based on spatial intersection.(3) Proposed a method of least squares surface fitting error points discrimination based on image segmentation. Starting from the object information put forward the error points discrimination method combination of image segmentation and surface fitting method, the method using the mean shift algorithm to segment the original image firstly, and then use different surface fitting function model for different object categories to judge the error points, in order to provide a more reliable data source for three-dimensional modeling of the city.(4) Analysis of mistaken points distribution characteristics and improvement on method of image dense matching. Based on the result of distinguishing error points,the collinear condition equation is used to project the error points onto the original image, and researching distribution characteristics of error points according to their location. On the basis of this, the method of image dense matching was improved, and the probability of error matching is reduced.