Fast Estimation of Relative Poses for 6-DOF Image Localization

The 6-DOF (Degrees Of Freedom) image localization, which aims to calculate the spatial position and rotation of a camera, is a challenging task for location-based services. In existing approaches, this problem is often tackled by finding the matches between 2D image points and 3D structure points so as to derive the location information using direct linear transformation (DLT). However, these approaches may fail to localize images when the 3D structure points are not available, especially for massive data. To address this problem, this paper presents a novel data-driven approach for 6-DOF image localization. In this approach, we propose to localize an image according to the position and rotation information of multiple similar images retrieved from a large reference dataset. From the reference images, a fast relative pose estimation algorithm is proposed to derive a set of candidate poses for the input image. Since each candidate pose actually encodes the relative rotation and direction of the input image with respect to a specific reference image, we can thus fuse all these candidate poses so that the 6-DOF location of the input image can be efficiently derived through least-square optimization. Experimental results show that our approach performs comparable with GPS devices in image localization. In addition, the proposed relative pose estimation algorithm is much faster than existing work.