Broad Area Stitching For Infrared Images Obtained By Airborne Scanning Imaging System

Airborne broad-area scanning large-field-of-view image stitching is to stitching the high spatial resolution image sequence acquired by an airborne infrared imager,so as to obtain aerial infrared images with large field of view and high spatial resolution.It is widely applied to military reconnaissance,disaster rescue fields.Differing from general image stitching,airborne broad-area scanning large-field image stitching is an orderly image stitching,whose imager provides servo information to aid image stitching.The core technologies of image stitching include image registration,bundle adjustment and blending.The traditional bundle adjustment method uses LM algorithm to solve for all of the camera parameters jointly.However,if the number of images needed to stitching is huge,LM algorithm takes too long to solve the camera parameters iteratively,even the iteration process does not converge.In addition,the traditional bundle adjustment method is not applicable to incremental splicing mode.To overcome the above problems,we improve the bundle adjustment method,and further develop a software to stitch airborne broad-area scanning infrared image with large field of view.Since the abundant images are needed to be processed in this task,the existing bundle adjustment methods always lead to long operation time and poor convergence.To address the problem,we propose a bundle adjustment method based on Alternating Direction Method of Multipliers(ADMM).Instead of LM algorithm,we exploit ADMM algorithm to solve the camera parameters of all images jointly,so that the quadratic sum of residual errors of all matched feature points is minimized.Compared with the traditional bundle adjustment method,the proposed method has better convergence and faster convergence speed in large-scale image stitching.Incremental stitching mode is also needed in airborne broad-area scanning stitching system,while bundle adjustment is to optimize the parameters of the joint camera for all stitching images,which is no longer applicable to the incremental stitching mode.In order to solve this problem,we propose an incremental camera parameter optimization method.This method just considers the local image registration information in the process of camera parameter optimization,and only adjusts the camera parameters of the new image at a time.The proposed method not only effectively solves the problem of bundling adjustment in incremental stitching,but also achieves high efficiency and excellent stitching results.Furthermore,we develop a software to stitch airborne broad-area scanning infrared image with large field of view.Extensive experiments are carried out on large-scale real-world infrared images,where the satisfactory results demonstrate the efficiency of airborne broad-area scanning large field of view infrared image stitching software.