Research On Panoramic Mosaic Using Dual Fisheye Images Based On Feature Matching

Panoramic vision,as an important part of virtual reality,is getting more and more attention.Panoramic mosaic is the core method of obtaining panorama.However,in the traditional multi-image-based panoramic mosaic method,the shooting process is complex and inefficient.To solve this problem,this thesis presents a panoramic mosaic method based on dual fisheye images.First,the fisheye images are mapped to the cylindrical panorama,and then the improved SIFT which is based on nonlinear diffusion filtering are used to detect feature points.Finally,the feature points are matched and purified,and the coincidence regions are merged.Experiments show that the method can be used to locate the spherical panorama with better quality to meet the engineering application requirements.The main work of this thesis includes the following three parts:(1)An improved SIFT method based on nonlinear diffusion filtering and multi-layer pyramid is proposed.Firstly,the nonlinear space is constructed.Then,the Hessian matrix is used to solve the feature points and the main direction is allocated.Finally,the feature is described by the multi-layer pyramid to produce the final descriptor.The algorithm can reduce the SIFT dimension and reduce the matching time without improving the matching precision,so as to improve the matching quality.(2)Based on the feature matching method,the overlapping area of the fisheye image is solved on the panoramic view of the cylinder.In this thesis,we first detect the contour of the fisheye image and map it into the cylindrical panoramic space.Then we perform the SIFT feature detection and matching in the cylindrical space to obtain the overlapping region.After image fusion,and then mapped to the spherical panoramic space,generate the final spherical panorama.(3)Implement a dual fisheye image of the panoramic mosaic system.We can use this system to convert dual fisheye images to a spherical panorama.In the meantime,we can roam the panorama with the interaction of mouse to zoom and rotate.