Automatic Upright Alignment of Multi-View Spherical Panoramas

Panorama images are now ubiquitously being used and are on their way to replace standard images. This makes it essential to adapt computer vision algorithms for panorama images. This is especially challenging for multi-view algorithms, as precise alignment of cameras is required to obtain panorama images facing a common direction. In this paper, we introduce an approach for automatic alignment of multiple panoramas. We extend a method for Upright Adjustment of spherical panoramas in order to include Rotational Alignment for a multi-view setup. Our technique is based on the most dominant vanishing points, which correlate in nearby panoramas. We conduct tests on one synthetic and several challenging real-world scenes with a multi-view setup. Additionally, we apply our algorithm to key frames of a real-world panorama video sequence for Rotational Video Stabilization. With only a few pixels of misalignment, we obtain visually pleasing results.