Research On Key Technologies Of Image Stitching

With the development of technology,the network bandwidth and the display equipment have been greatly improved.Along with this,image acquisition equipment is also pursuing a larger resolution and a larger field of view image,as a result,there are more and more high-definition and wide-angle cameras.Hardware devices,such as wide-angle cameras and cameras array,can capture wide-angle images,however,this kind of equipment is expensive,and not suitable for some special scenes.Image stitching is a kind of method that combines multiple images into an image which has a larger field of view,so using image stitching to get wide-angle and large field of view images is more and more important.Image stitching consists of two main stages,image registration stage and image composition stage.In the first stage,single or multiple homography matrices are obtained,then in the second stage,in order to get mosaic image,multiple images with overlapping regions are registered together with the minimum structural distortion by homography matrices.In the image registration stage,there are many algorithms to get the registration parameters between images,and in order to better understand a kind of image stitching method based on 3D reconstruction,in the practical application of multi-degree-of-freedom and multi-scale spatial modeling project of airport scene,by using some special optimization techniques,a more accurate scene is constructed.In addition,for parallax image stitching which is difficult to stitch by existing algorithms,by analyzing the causes of parallax,a local registration method is proposed.The main work and ideas are as follows:1.In the video stitching process,in order to improve the efficiency of 4K video stitching,by sharing registration and post-processing parameters within a certain period of time can greatly improve the efficiency of video stitching.In addition,in the stage of image registration,the feature extraction and matching speed can be greatly improved by drawing ROIs in images,what's more,drawing ROIs in images can make the registration more focus on the airport runway scene,so that it can reducing the structure distortion of the mosaic,and improving stitching performance.2.In the pre-processing stage,using the Shearlet transformation to remove the noise of images,after that,the accuracy of image feature extraction is increased,and the accuracy of multi-degree-of-freedom and multi-scale spatial modeling of airport scene is improved.3.When the distance of camera baseline is large and the scene has multiple depth of field,the image parallax is obvious.In order to solve this problem,a ‘multi-layer' image stitching method,‘a depth-based adaptive position compensation parallax image stitching method',is proposed.Compared with the conventional methods,the feasibility of the theory and algorithm is proved.