ARC-Loc: Leveraging Azimuthal Ray Convergence as a Geometric Cue for Direct Cross-View Localization

The researchers proposed a new method called ARC-Loc, which aims to use azimuth rays convergence as a geometric clue for direct cross-view positioning. This method converts ground关键点 into azimuth rays on satellite images and utilizes the characteristic that these rays ideally converge at the user’s location. A minimization of azimuth ray convergence (ARC) solver and loss function are introduced to identify intersection points and optimize the matching network. ARC-Loc eliminates the reliance on computationally intensive bird’s-eye view transformations and external depth base models, enabling direct ground-to-satellite image matching and positioning without an external depth model. Experiments on the VIGOR and KITTI datasets show that this method maintains competitive positioning accuracy compared to recent methods.