Adaptive Cylindrical Panoramas Mosaic
自适应柱状全景图拼接

In this paper,the problem of efficiently and precisely constructing the cylindrical panoramas view of a scene from an image sequence is considered.The proposed method firstly detects correspondences based on gray level edge feature so that the focal lens can be counted based on correspondences uniqueness.Then the focal lens will be directly used in following image matching processing with a novel adaptive mixed gray level image matching algorithm and outputs a rough panorama result.The width of it reacts to a judgment to adjust the focal lens,which induces an iterative computation to guarantee an exact focal lens.Finally an angle-oriented one-dimension projection template matching algorithm will be employed to modify the correspondences in narrow range and the final panorama is obtained by linear image blending.Adaptive using multiple image matching algorithms according to self-relation variant after precise focal lens searching,the method redraws the accurate overlapping image horizontally and vertically.Experimental results proved that the proposed approach are given works well over most image sequence capture by handholding camera,and is fast,accurate and robust enough for real time applications.