Research On Aerial Image Stitching Technology Based On KAZE And GMS

In recent years,aerial photography technology has gradually matured.It has been widely used in danger area investigation,ground mapping,disaster monitoring and evaluation,marine environmental monitoring and so on.The coverage area of a single aerial image is too small.In order to display more comprehensive information,the aerial images need to be stitched.Aerial image is susceptible to the changes of light,rotation and scale,and has the characteristics of high resolution and large scale.It is required that the stitching algorithm has better matching performance while maintaining a certain matching speed.Feature-based image stitching technology can deal with the above problems and be used to stitch aerial images well.Local invariant feature extraction and feature matching are two key technologies in feature-based aerial image stitching algorithm.In order to obtain accurate aerial stitching images efficiently and quickly,this paper focuses on the feature point extraction algorithm and feature matching algorithm for aerial images.In terms of feature point detection and description,the calculation speed of KAZE algorithm is slow.Therefore,a feature detection and description algorithm based on improved KAZE is proposed in this paper.In order to improve the calculation speed while preserving the KAZE algorithm's better matching performance,the fast explicit diffusion is used to construct the non-linear scale space,and fast retina keypoint algorithm is adopted to describe the feature points in improved KAZE.In order to verify the effectiveness of the proposed algorithm,we combine the improved KAZE algorithm with the k-Nearest Neighbor algorithm which introduces the constraint of motion smoothness for aerial image registration.Simulation experiments based on Mikolajczyk and Schmid datasets and aerial images verify that the proposed algorithm has better illumination invariance,rotation invariance and scale invariance while effectively improving the speed of feature point extraction and matching,and it can be used to detect and describe the feature of aerial image well.In terms of feature matching,the registration performance of grid-based motion statistical feature matching algorithm needs to be improved.Therefore,an improved grid-based motion statistics aerial image feature matching algorithm is proposed in this paper.The improved feature matching algorithm based on grid-based motion statistics establishs the corresponding relationship of feature points based on bidirectional brute force matching algorithm after extracting feature points of the image.On the basis,the grid-based motion statistics algorithm is used to separate true matches and false matches,and the matching results are subjected to epipolar constraints to obtain higher-quality matching pairs.It improves the matching performance while retaining the higher matching speed of the grid-based motion statistics feature matching algorithm.Simulation experiments based on aerial images show that the proposed algorithm can greatly the matching performance under the changing of light,rotation and scale while maintaining high matching speed and robustness and can better match the aerial image feature.Finally,we combine the improved KAZE algorithm with the improved grid-based motion statistics feature matching algorithm for the stitching of aerial images.Simulation experiments based on aerial images show that the aerial image stitched by the proposed algorithm has no stitching,no ghosting,and is overly natural in the overlapping area.The proposed algorithm can stitch aerial images effectively.