Research On Parallax Optimization Of Remote Sensing Images Based On Multi-model Fitting

In order to realize the extraction and mining of visual data such as images and videos,computer vision has gradually developed into an independent and mature discipline.Three-dimensional reconstruction also plays an important role in many branches of science,and multi-model fitting belongs to a large class of algorithms in the statistical category.Its application field is not restricted to computer vision or even computer science.Its high algorithm is robust.In addition to eliminating outliers,the model fits the sample to a more appropriate model.In the binocular stereo vision system for remote sensing images,the parallax calculation is related to the subsequent elevation calculation,which directly affects the accuracy and accuracy of the final reconstruction result.However,considering the input two remote sensing images after changing the viewing angle of the satellite,the occlusion areas appearing in the images,the lighting conditions change,the distortion of the light diffraction by the atmosphere and the expansion of the foreground,etc.Less than some corresponding points on the left,the parallax map generates a null value area.For this phenomenon,this paper attempts to combine the theory of multi-view geometry with the idea of multi-model fitting algorithm to estimate the approximate homography transformation matrix of the pixels in the discontinuous discontinuous area,so as to calculate the coordinates of the matching point of the point,and then estimate the parallax of the point In the process of optimizing the parallax,the multi-model fitting algorithm was selected and improved in a targeted manner,which not only improved the efficiency but also improved the encapsulation and usability of the program.Finally,the feature vectors of the two point sets of the point to be estimated and the calculated matching point of this point are calculated and sampled to match.While quantifying the reliability of the parallax estimation reliability,it is also compared with the parallax estimation using traditional methods such as local gray histograms.The results are compared and the feasibility of the method is demonstrated in three dimensions.The main work of this article is as follows:1.Investigate the current research status of 3D reconstruction of remote sensing images at home and abroad,summarize the similarities and differences between different methods,and find a series of difficulties and pain points in applied algorithms.At the same time,the multi-model fitting algorithm is introduced into categories,and the possibility of the two working together is explained in combination with 3D reconstruction.2.The s2p satellite remote sensing image processing stream is developed for the applicability of the project in this article.The limit correction method is used in the preprocessing stage to limit the parallax in one direction with the minimum error,and the image is fused in the cost calculation stage.The second-order derivative of sigma further enriches the cost function and improves the accuracy of stereo matching3.Inspired by the J-Linkage algorithm,the Seq-RANSAC algorithm was improved by local initialization of the model,and the effect of model fitting was improved without affecting the operation efficiency of the algorithm.In the disparity classification stage,the method of cross search and sliding window is used to effectively alleviate the single problem of disparity classification in a large null area4.Use the project image to test the algorithm in this paper,and carry out comparative experiments in horizontal and vertical directions,and use a SURF descriptor-based evaluation index to quantitatively analyze the pros and cons of the algorithm.