Research On Remote Sensing Image Registration Method Of Unmanned Aerial Vehicle Based On Point Feature

Image mosaic refers to the registration and fusion of two or more images and overlapping regions to produce images with higher viewing angle and resolution.Because of the low demand for acquisition equipment,it has been widely used in such areas as space remote sensing,military surveillance,medical imaging and virtual reality.Image registration is the key and difficult problem of image mosaic.The scanning invariant feature transformation(SIFT)algorithm is widely used in high resolution images,but when the pixel gray value of a remote sensing image has an obvious nonlinear difference,the existing method is difficult to register.Therefore,two improved algorithms based on point features are proposed for the characteristics of remote sensing images of Unmanned Aerial Vehicle:In view of the problems of the position scale orientation SIFT(PSO-SIFT)algorithm in remote sensing images registration of Unmanned Aerial Vehicle with fewer correct matching points,lower correct matching rate,and long running time,this thesis proposes an enhanced image registration algorithm based on modified PSO-SIFT.First of all,the idea of"back'' type block is adopted to describe the characteristics.The experimental results show that the rectangular building area with more diagonal points fits well.Then,three matching algorithms are combined to carry out feature matching.Experimental results show that the matching strategy not only ensures the accuracy of matching but also increases the number of correct matching pairs.Aiming at the problem that the remote sensing image registration algorithm based on anisotropic diffusion and Harris is directly applied to the remote sensing image of Unmanned Aerial Vehicle,which leads to the problem that the number of correctly matched feature point pairs is small,especially the time is particularly long,this thesis proposes a remote sensing image registration algorithm based on anisotropic scale space.Firstly,the adaptive side window filtering technology is used to generate anisotropic scale space and keep the edge of the image.Then the improved neighborhood block idea is used to reduce the dimension of the feature description vector.Finally,two combinatorial algorithms are used to match,more matching points are obtained,and the registration accuracy is guaranteed.Experimental results show that the algorithm is superior to the original algorithm in the number,accuracy and time of matching pairs.