Research Of Aerial Image Stiching Method Based On Grid Optimizationation

The aerial image stitching method is to take a series of related images which are taken by the aerial cameras such as drones,through feature points extracting,matching and selecting?image registration and image fusion,finally stitching to a large angle image containing all scenes and targets,the main application is in digital map generation,drone navigation,target tracking and identification,natural disaster prevention and controlling.This paper introduces a complete aerial image stitching process,using grid adjustment and optimization to reduce mismatch and image distortion,improved image registration accuracy,balanced the brightness difference between images during the fusion process,and achieve a smooth transition of overlapping regions.A higher quality aerial image stitching result is obtained.The specific work of this paper is as follows:(1)For the mismatching of feature points,Using the Learned Invariant Feature Transform feature point model to get more feature point matching information,through the Random Sample Consensus algorithm screening to get more accurate feature point pairs,and improving the quality of the generated control points,giving the improved method of generating control points in mesh transformation.For the image gridding error in the registration process,the control point combining with the least squares mesh constraint is gived to adjust the image mesh.The accuracy of image registration and splicing is improved by grid optimization.(2)In order to maximize the original image information for the stitching results,this paper uses the improved Shape-Preserving Half-Projective image stitching algorithm.The algorithm divides the image into different regions and preserves the original image details through similarity image transformation.However,the error of the global homography transformation is large used by the algorithm in the registration process.Combined with the least squares mesh can improve the registration process of the algorithm,improving the accuracy of the algorithm in local registration,and the scene alignment of the local image region is further improved.The Shape-Preserving Half-Projective algorithm is still used in the global image projection for weighted linear fusion.Finally,a brightness balancing method is adopted to eliminate the influence of brightness difference between images on the stitching result,so it can obtain more accurate aerial image stitching results.The aerial image data in this paper is derived from the aerial image of the drone and the online algorithm data sets,the aerial image data is classified according to the height.Thought two and multiple images' stitching experiments,the experimental results are compared with other related methods,which is proving the superiority of our paper's stitching algorithm.