Research On UAV Image Automatic Mosaic Technology

The UAV aerial image mosaic technology which mosaic a group of content-related aerial image into a big image automatically by image registration and image fusion technology,and make the mosaicked result cover the whole shooting scene content with large view and high resolution,it has important application value in the digital map and virtual environment generation,target tracking and UAV navigation,military reconnaissance air monitoring and early warning,disaster control etc..This thesis,along the process of UAV aerial image automatic mosaic,studied in depth the key technologies involved in aerial images mosaic,which mainly including image registration,parameter estimation and optimization,and image fusion.Image registration is the basis of image splicing and has a great influence on the speed of image mosaic and the accuracy of mosaicked results.In this thesis,a feature extraction method based on Log-Gabor transform is studied for images with simple texture features,and the registration effects acquired by SIFT and Log-Gabor methods on the regions of deserts and grasslands is experimentally compared.Secondly,a multi-image registration strategy based on adjacent relation determination and registration is proposed in this thesis.This method extracts the feature by solving the maximum extremum stable region of the image and describes the region extracted using the SURF description operator,preliminary determines the interaction between the images through coarse registration,and obtains the pair of feature points by the accurate registration of adjacent images.Because the region-based feature matching algorithm has a low complexity and gets a small number of extracted feature points,the interaction between images can be determined quickly,and the accurate registration between non-adjacent images can be avoided,which results in the time required for registration is greatly reduced.In this thesis,the camera imaging model in UAV aerial photography is detailed analyzed from the perspective of computer vision,and a simple and efficient 6-DOF camera imaging model is deduced.According to the causes of errors in the multiple aerial images mosaic,a parametric optimization model with constraints is proposed.And through several groups of comparison experiments,the feasibility of the proposed 6-DOF imaging model and parameter optimization adjustment method with constraints is verified.Seamline-based image fusion is prone to local discontinuous on both sides of the seamline.In response to this phenomenon,this thesis first solves the minimum connected sub-region,and then recursive backtracking search for seamline within this connected region.This method eliminates the ‘misalignment' phenomenon visually and greatly improves the timeliness.Secondly,according to the characteristics of UAV aerial images mosaic,this thesis applies the idea of minimum spanning tree to search an optimal path as the seamline.Through several sets of image stitching experiments and compares with the dynamic programming method which makes the maximum mismatch minimized,the results prove that this method has a good mosaicked effect.