Research On The Key Technology Of UAV Aerial Image Mosaic

Unmanned Aerial Vehicle(UAV)aerial as a new means of remote sensing,with timeliness,low cost,flexible fast,high resolution and other advantages,more and more widely used.But in the process of UAV aerial flight by the flight height,camera focal length and other restrictions,the acquisition of a single shot images are often difficult to have a comprehensive understanding of the target area,so to carry out a number of aerial images automatic stitching technology has a strong practical significance This thesis mainly focuses on the key technology of automatic stitching of UAV aerial images.The main work is as follows:In the aspect of automatic sequencing of aerial image sequences,the phase correlation method of phase correlation method is introduced into the sorting algorithm,and the maximum correlation criterion and the adjacent image position are studied.The principle of relationship judgment is effective to solve the problem of sorting between aerial image sequences,such as translation,rotation and scale scaling,which avoids manual intervention and enhances the application of the algorithm.In the aspect of aerial image registration technology research,based on the characteristics of Scale Invariant Feature Transform(SIFT)algorithm,SIFT algorithm is used to extract features,and Best Bin First(BBF)optimal node priority algorithm is used to realize the initial matching of feature points.Random Sample Consensus(RANSAC)algorithm is used to purify feature matching.On the basis of this,the deficiency of SIFT algorithm is complex and the lack of real-time is improved in both feature extraction and feature matching.In the aspect of feature extraction,we study a unique SIFT feature description sub-construction method,select the ring as a neighborhood to construct the key descriptor,use the adaptive quantization strategy to divide the local region and calculate the gradient histogram,and finally form 96-dimension instead of 128-dimensional feature description vector.In terms of feature matching: the RANSAC algorithm is based on the different characteristics of the image to give an adaptive distance ratio threshold selection method,which eliminates the decrease of the robustness of the feature descriptor due to the decrease of the robustness of the feature descriptor.In the aspect of feature matching,The problem of reduced accuracy.The improved SIFT algorithm reduces the complexity of the SIFT registration algorithm to a certain extent,and ensures that the robustness and matching rate of the characteristic operator satisfy certain requirements.In the aspect of aerial image fusion and splicing,a local brightness adjustment method is introduced for the weighted average fusion algorithm.The gray value is used to adjust the brightness value of the overlapping region,and the splicing image caused by large exposure difference is effectively solved.Differences in visual effects.Finally,the whole transformation model between images is established,and the correspondence between images is described by using the perspective transformation model matrix(H),and the fast splicing of multiple aerial images is realized.