Improvement on Image Rotation for Relative Self-Localization Estimation

There are many methods to localize its position based on visual sensing schemes in indoor environment. This paper presents the problem of finding the correspondences of images feature’s descriptors when the images have large rotation. SIFT and SURF have always been considered as very effective algorithms to extract interest points and their orientation and descriptors. For descriptors, one of both uses a lot of time to calculate descriptors and the other has not good performance in large rotation of the image. In this paper, we propose an improved algorithm to calculate interest points’ descriptors for relative self-localization estimation. The proposed algorithm will satisfy descriptor invariant when the image rotates. Meanwhile, the proposed method reduces the calculated time as much as possible. Interest point’s descriptors are formed by resampling local regions.