Pan360: INS Assisted 360-Degree Panorama

We propose a practical stitching method in dealing with input images with weak salient points, high luminance variations, and repetitive structures in an artificial environment. We address these panorama-stitching challenges by inferring orientation of capturing pose by inertial sensors inside a mobile device. Our contributions include identifying an effective way to consider orientation, providing novel methods to further improve feature-matching rate, and solving the vignette effect in low light environments. Experimental results demonstrate that our global alignment method outperforms other stitchers, and the color tone is well maintained and more appealing.