Beyond Matching to Tiles: Bridging Unaligned Aerial and Satellite Views for Vision-Only UAV Navigation

TL;DR

Bearing-UAV introduces a vision-only cross-view navigation approach that jointly predicts UAV location and heading, improving robustness and accuracy over existing methods, and provides a new benchmark dataset for evaluation.

Contribution

The paper proposes Bearing-UAV, a novel method for cross-view UAV navigation that encodes spatial relationships and jointly predicts location and heading, enhancing robustness and efficiency.

Findings

Bearing-UAV achieves lower localization error than previous methods.

The approach is robust to cross-view variations and feature-sparse conditions.

The new benchmark dataset facilitates evaluation of cross-view localization methods.

Abstract

Recent advances in cross-view geo-localization (CVGL) methods have shown strong potential for supporting unmanned aerial vehicle (UAV) navigation in GNSS-denied environments. However, existing work predominantly focuses on matching UAV views to onboard map tiles, which introduces an inherent trade-off between accuracy and storage overhead, and overlooks the importance of the UAV's heading during navigation. Moreover, the substantial discrepancies and varying overlaps in cross-view scenarios have been insufficiently considered, limiting their generalization to real-world scenarios. In this paper, we present Bearing-UAV, a purely vision-driven cross-view navigation method that jointly predicts UAV absolute location and heading from neighboring features, enabling accurate, lightweight, and robust navigation in the wild. Our method leverages global and local structural features and explicitly encodes relative spatial relationships, making it robust to cross-view variations, misalignment, and feature-sparse conditions. We also present Bearing-UAV-90k, a multi-city benchmark for evaluating cross-view localization and navigation. Extensive experiments show encouraging results that Bearing-UAV yields lower localization error than previous matching/retrieval paradigm across diverse terrains. Our code and dataset will be made publicly available.