| As one of the hot topics in the field of image graphics in recent years,virtual reality（VR）means to provide a virtual 3D surrounding space where can give users immersive experience by making full use of humans’senses such as vision on hearing and touch.A crucial step in VR’s preparing work is taking advantage of image stitching technique to form a panorama.Panoramic stitching technique is point to synthesis more than one images made at the same time or different times about a scene and generate a new complete and coherent explanation about this scenario.Panoramic stitching technology mainly involves three modules including image calibration,image matching and image fusion.This paper explains and analyzes this technology focusing on the three parts,and then give the corresponding improvement algorithms.In the last,this paper compared and summarized these techniques.In the part of image calibration,the existing ordinary camera’s visual field is so small that it doesn’t fit for panoramic stitching technology because it can’t extract all scene images in vision range in a short time.On the contrary,fisheye camera,with the advantages of short focal length and great view angle,a lot of the fisheye cameras’viewpoint are more than 120o or 180o,which can be used to gather images within the scope of vision horizontal or vertical direction quickly.However,fisheye image’s distortion is so serious that need effective correction algorithms to do processing.This paper introduces several differentfisheye camera projection models,effective area extracting algorithms and several correction algorithms including latitude and longitude mapping correction,double longitude coordinates calibration and correction based on Scaramuzza.Finally,this paper puts forward the improved algorithm based on longitude coordinates correction and compares the advantages and disadvantages of several kinds of fisheye image correction algorithms.In the part of image matching and stitching,according to the path of feature points detection,matching,denoising process and H transformation matrix,this paper respectively introduces Harris,SIFT feature point detection algorithm and SIFT feature point matching algorithm,histogram feature point matching algorithm primarily,and then introduces the improvement algorithm based on PCA-PSO.At the same time,the paper compares these matching algorithms.On this basis,this paper also introduces the RANSAC feature point denoising and H transformation matrix procedure.In the images fusion part,this paper introduces some commonly fusion algorithms including the classic gradually linear fusion algorithm,the best suture fusion algorithm which can solve the ghosting problem and more novel poisson fusion algorithm.In the last,the paper puts forward the improved algorithm based on the gradually linear fusion algorithm.According to the processing of fisheye image correction,image matching and stitching,image fusion,this paper provides a comprehensive description of panoramic stitching technology. |
