High-precision Three-dimensional Reconstruction Of Remote Sensing Images Based On Multiple-model Fitting

Remote sensing technology is widely used in more and more fields due to its advantages such as large detection range,fast data acquisition speed,and limited conditions.Three-dimensional reconstruction technology is widely used in urban planning,disaster detection,military reconnaissance,and precision weapon strikes,and has very important research significance.This thesis implements a high-precision three-dimensional reconstruction technology based on multi-model fitting.It mainly studies plane estimation based on multi-model fitting,plane fusion based on dissimilarity coefficients,and three-dimensional reconstruction based on rational function models.The inspection data obtained by stereo matching is optimized to improve the effect of three-dimensional reconstruction.This thesis proposes a multi-model fitting and plane fusion plane estimation method based on semi-global stereo matching.Correcting and filling the parallax in the plane improves the integrity and accuracy of the parallax.On this basis,three-dimensional reconstruction is performed,which improves the effect of three-dimensional reconstruction.The algorithm is divided into the following three steps.First,the method combines superpixel segmentation and affine model to form a multi-model fitting method to estimate the sub-plane parameters of the remote sensing image.The image is divided into sub-regions by a super-pixel segmentation method based on the entropy rate,and the planes in the sub-regions are estimated by a weighted iterative least square method based on M-estimation.This method is more robust than the least squares method,and the model fitted by this method is more accurate.Besides,an outlier detection algorithm based on spatial topology is used to fit the plane coefficients,which is used to eliminate the influence of outliers on the plane fitting process and improve the fitting accuracy.Secondly,this thesis implements a plane fusion algorithm based on dissimilarity coefficients,and optimizes the parallax data obtained from stereo matching based on the final fusion result image.This thesis adopts the idea of object-based region fusion,and proposes an improved merging cost function based on affine model.The fusion of sub-planes is realized through the merging strategy based on dissimilarity coefficients.After completing the plane fusion,based on the affine model of each plane after the fusion,fill in the missing matching points and remove the outliers to complete the parallax optimization of each plane.Finally,the three-dimensional reconstruction is completed through the rational function model.This thesis analyzes and studies the principle of elevation solution based on the rational function model,establishes a rational function model by mapping the image coordinates and geographic coordinates,and calculates the elevation data based on the parallax data combined with the model to complete the three-dimensional reconstruction.In summary,the three-dimensional reconstruction method based on multi-model fitting proposed in this thesis fits the plane model to different planes separated by plane fusion,and completes parallax optimization based on the fitted plane.The experimental results prove that this method improves the matching rate of remote sensing image pairs and improves the integrity and accuracy of the elevation data results.