Automatic spherical panorama generation with two fisheye images

Panoramas are often used to enhance immersive experiences. Former panorama generation methods with fisheye images either require pre-calibrating the rotation of fisheye cameras with a calibration rig or manually selecting correspondences between images. In this paper, we propose an automatic spherical panorama generation method with two fisheye images. The method contains fisheye camera calibration and panorama stitching. In the calibration, the field of view and circular boundary are used to initialize the focal length, and line perspective constraint on sphere is used to refine the intrinsic parameters. In the panorama stitching, we use scale invariant feature transform (SIFT) features and RANSAC to get the fisheye correspondences. Then the correspondences are used to refine the estimation of rotation and intrinsic parameters. Finally, a spherical panorama is obtained by warping the fisheye images. In the panorama stitching process, no user input is needed. Experimental results on both simulated and real images show that our method is effective.