Research On High-resolution Panoramic Image Stitching Algorithm

Panoramic images can bring unparalleled visual experience to people.Using panoramic image stitching algorithms to stitch images is the main way to obtain panoramic images.Panoramic image stitching algorithms iswidely appliedin security monitoring,satellite remote sensing,virtual reality,electronic maps,etc.This paper designs and implements a 360-degree panoramic image real-time stitching system.Based on the existing image stitching algorithms,This paper propose the optimization methodson the registration accuracy of the panoramic image stitching algorithm,color consistency and real-timestitching.The panoramic image real-time stitching system is consisted of a panoramic camera,a switch,and a computer.The design of the panoramic camera adopts a modular design concept.The camera is composed of eight network camera modules,four micro switches,power supply circuits and cooling fans.The panoramic camera can capture video in eight directions in real time and push it to the computer which in the same local area network.The functions of storing,stitching and displaying images are completed on the computer.This paper proposes an optimization algorithm for grid transformation,which improves the accuracy of image registration and reduces the problem of ghosting in the stitched imageseffectively.The sub-grid similarity optimization term,sub-grid homography optimization term,and overall mesh similarity optimization term are used to adjust the vertex position of the mesh to make the image registration accuracy based on the local homography transformation of the mesh.the stitching result is more natural with the method.A color correction algorithm for panoramic images based on RGB histogram is proposed,which solves the problem of color imbalance of panoramic images obtained by stitching.In order to make full use of the results of the image registration algorithm to accurately extract the overlapped regions between adjacent images,and use the image color features of the overlapped regions as the basis for migration,to achieve the accurate migration of image color features.This paper proposes a real-time stitching solution for panoramic images based on multi-threading technology and GPU acceleration,which effectively improves the real-time performance of panoramic image stitching.The feasibility and acceleration effect of GPU parallel computing are proved by experimental analysis,and the parallel image stitchingalgorithm is suitable for parallelization.Combined with the panoramic image stitchingsystem designed in this paper,the multi-threaded technology is used to simultaneously decode eight-channel 1080 p high-definition video,and the GPU is used to implement real-time correction of the image.The image transformation parameters calculated by the stitchingalgorithm are updated in the GPUdynamically.The GPU uses image transformation parameters to transform and fuse the corrected image.Because the video image does not change significantly in a short time,the speed of the update of the transformation parameters has little effect on the mosaic image stitching results.The experimental results show that the real-time stitching solution of panoramic images has a good stitching effect,which basically meets the needs of real-time stitching of panoramic images.