Spherical panorama stitching based on feature matching and optical flow

Abstract

In recent yeas, 360° spherical panorama images and videos have seen huge adoption in virtual reality research. Image mosaic is the main technology to stitch the little scale images to a large scale panoramic image. Because of the gross distortion in the edge of fisheye lens, the misregistration on the corrected images leads to obvious ghosting after stitched. In this paper, we use pyramid LK optical flow algorithm to reduce the misregistration areas by remapping with interpolation algorithm. Additionally, we use sample point algorithm to adjust the brightness of images for minimize visual seams. Experiments show convincing evidence that the effect of our spherical panorama stitching improves significantly.