Motion Estimation And Stereo Matching Of Urban Remote Sensing Images

With the continuous development of remote sensing,the acquisition of high-resolution remote sensing image information has become more and more simple and quick,and in the field of people's livelihood,military,disaster prevention,urban planning and other fields,it also increasingly relies on the acquisition of spatial information.As the basic part of the field of computer vision,stereo matching has a direct impact on post processing,for example target recognition and 3D reconstruction.Stereo matching attracts many experts' research,and new matching algorithms appear every year.However,most of these matching methods are concerned with the images taken by an ordinary camera that conforms to the epipolar geometry.But the imaging geometry of the linear array CCD sensor on the high resolution remote sensing satellite doesn't satisfy the epipolar geometry.In addition,the information of remote sensing images is more complex and more interfering than ordinary images,which all increase the difficulty of matching the real remote sensing images.Many stereo matching algorithms don't suit remote sensing images,and the research focused on stereo matching of remote sensing images is relatively less.This paper focuses on the stereo matching of urban remote sensing images,and puts forward a new matching framework.The main research work of this paper is as follows:(1)First registrate the image,so as to minimize the search space for subsequent matching.The feature points are extracted and matched by SURF,and the matching points set which contains mismatched points is obtained.The resulting set of points is processed using the MLESAC algorithm to obtain an accurate set of interior points.Then using the DLT algorithm,the affine transformation matrix between two images is calculated by using the correct matching points.The right graph is processed with this matrix to get the rough rectified image.(2)Because the registration image still has two-dimensional disparity,the motion estimation algorithm based on block matching is used to calculate the displacement of the corresponding subblock.Because the artificial object of the urban remote sensing image has a certain distribution characteristic,like the same building roof or the same floor area of the same buildings,obeying the same movement,therefore,the method based on Mean Shift is used to gather the object of the displacement into different classes.Use MRF to deal with the clustering results,and the center point of the sub block is approximated as the matching point.For further precise matching results,a local stereo matching method based on MI is applied to search around the approximate matching points to get the accurate matching feature points.(3)In order to prove the effectiveness of the proposed algorithm flow,the simulated urban remote sensing images and real urban remote sensing images are processed respectively.For the simulated image,the images have already after epipolar line rectification,we can use the disparity to calculate altitude,and compare with the given accurate altitude information.For real images,the artificial observation is the only method to judge the matching situation.The experimental results show that the proposed method can obtain ideal results on both the simulated and real images,which proves the effectiveness of the proposed scheme.