Research On Stereo Matching Of Remote Sensing Image Based On Affine Invariant Features

In the filed of computer binocular vision,stereo matching of remote sensing images has always been a challenging task.At present,with the continuous development and maturity of related technologies in the filed of satellite remote sensing,it has been very quick and convenient to obtain high-resolution urban remote sensing images.Stereo matching is a technique for reconstructing and sensing a three-dimensional world using two-dimensional images,however,due to noise interference,lens distortion and light effects that may occur during the shooting process,the complex information carried in remote sensing images may change,at the same time,due to the limitations of the edges of buildings in urban remote sensing images,the final stereo matching accuracy will be affected.There is no universal stereo matching algorithm to obtain accurate matching results,therefore,the stereo matching research of remote sensing images has very important theoretical value and practical significance.The research work of this thesis includes the following three aspects:(1)This thesis proposes the affine invariant matching by using sub-pixel edge information,aiming at the problem of inaccurate extraction of information on edge points of buildings in remote sensing images.This method accurately limits the traditional pixel precision to the fractional range,first the image is preprocessed,and then the subpixel edge extraction is performed on the remote sensing image by the subpixel edge detection algorithm based on the edge and the area effect obtained from the edge model,subdivide integer-level pixels,extracting the position of the feature points where the gradients on both sides change significantly,making location information more accurate.The method is robust to noise interference in the image and is not interfered by the ground information in the image,which ensures the accuracy of feature point extraction.(2)In order to obtain a more accurate match,this thesis proposes a global stereo matching algorithm based on edge detection.The integer-level edge points extracted by the Canny algorithm can be matched by using the disparity information,however due to the inaccuracy of Canny edge extraction and integer level information,mismatching occurs.Therefore,this thesis maps the integer-level points to the sub-pixel point information.If there is only one sub-pixel point information in the range of an integer level,the sub-pixel matching can be improved according to the Canny edge detection operator.(3)In order to further demonstrate the effectiveness of the proposed algorithm,this thesis selects real remote sensing images and simulated remote sensing images for testing experiments.By calculating the absolute elevation error of the building and using the subpixel matching algorithm based on super resolution reconstruction as a comparison,it can be shown that the proposed algorithm can achieve sub-pixel precision in the elevation calculation,and the matching rate is better than the compared algorithm,and a more accurate matching effect is obtained.