Organizations: School of Electronic Information, Wuhan University, Wuhan 430072, China · School of Artificial Intelligence, Wuhan University, Wuhan 430072, China · Technology and Engineering Center for Space Utilization and the Key Laboratory of Space Utilization, Chinese Academy of Sciences, Beijing 100094, China · School of Aeronautics and Astronautics, University of Chinese Academy of Sciences, Beijing 100049, China
Cross-view geo-localization for Unmanned Aerial Vehicles (UAVs) operating in GNSS-denied environments remains challenging due to the severe geometric discrepancy between oblique UAV imagery and orthogonal satellite maps. Most existing methods address this problem through a decoupled pipeline of place retrieval and pose estimation, implicitly treating perspective distortion as appearance noise rather than an explicit geometric transformation. In this work, we propose a geometry-aware UAV geo-localization framework that explicitly models the 3D scene geometry to unify coarse place recognition and fine-grained pose estimation within a single inference pipeline. Our approach reconstructs a local 3D scene from multi-view UAV image sequences using a Visual Geometry Grounded Transformer (VGGT), and renders a virtual Bird's-Eye View (BEV) representation that orthorectifies the UAV perspective to align with satellite imagery. This BEV serves as a geometric intermediary that enables robust cross-view retrieval and provides spatial priors for accurate 3 Degrees of Freedom (3-DoF) pose regression. To efficiently handle multiple location hypotheses, we introduce a Satellite-wise Attention Block that isolates the interaction between each satellite candidate and the reconstructed UAV scene, preventing inter-candidate interference while maintaining linear computational complexity. In addition, we release a recalibrated version of the University-1652 dataset with precise coordinate annotations and spatial overlap analysis, enabling rigorous evaluation of end-to-end localization accuracy. Extensive experiments on the refined University-1652 benchmark and SUES-200 demonstrate that our method significantly outperforms state-of-the-art baselines, achieving robust meter-level localization accuracy and improved generalization in complex urban environments.
Figures & tables
Figure 1: Framework Comparison. We show the comparison with existing two-stage UAV geo-localization methods and our unified UAV geo-localization framework.
Figure 2: Overview of the unified UAV geo-localization pipeline. Given a sequence of oblique UAV query images, the pipeline extracts the representation using VGGT to reconstruct a local 3D scene. Based on this shared representation, a satellite-aligned BEV is rendered to support coarse-level satellite retrieval. The retrieved satellite candidates are then independently aligned with the reconstructed 3D scene within the same feature space, enabling fine-grained regression of the absolute 3-DoF UAV pose (GPS coordinates and heading).
Figure 3: Satellite-wise Attention Block. The proposed satellite-wise attention block lets the satellite tokens attend to the UAV tokens independently, followed by the candidate validations.
Figure 4: Similarity Score of Features.
Figure 5: Geometric Relationship between UAV Camera and Satellite Tiles. Visualization of the University-Pose dataset geometry. The red arrow represents the 3-DoF UAV pose. Notably, the dataset contains spatially overlapped satellite tiles, providing dense candidates that facilitate robust pose estimation.
Figure 6: Overlapped Satellite Images Statistic.
Methods
University-1652
SUES-200 (150m)
R@1 ↑
R@5 ↑
R@10 ↑
AP ↑
R IoU @1 ↑
R IoU @5 ↑
Time(s) ↓
R@1 ↑
AP ↑
University ( Zheng et al., 2020 )
69.3
86.7
91.2
73.1
-
-
0.8
67.8
72.7
AnyLoc ( Keetha et al., 2023 )
36.2
59.1
57.3
36.2
37.2
60.2
1.9
30.0
42.7
SegVLAD ( Garg et al., 2024 )
56.8
68.1
76.9
56.9
60.2
69.3
16.1
56.0
59.1
UCVGL ( Li et al., 2025 )
54.5
63.2
77.9
66.2
57.1
74.1
62.3
56.0
60.5
Ours (Initial BEV)
58.8
78.7
84.2
63.2
63.2
79.2
3.5
56.5
61.6
Table 1: Retrieval Result on University-1652 and SUES-200.
IoU Threshold τ
R IoU@1↑
R IoU@5↑
R IoU@10↑
0.1
84.9
89.2
89.9
0.2
83.9
88.4
89.4
0.5
80.6
88.3
89.4
0.8
79.9
88.2
89.3
1.0
79.0
88.2
89.3
Table 2: IoU-based Metric on Retrieval Results.
Figure 7: Visualization of the BEV Refinement.
Refine Time
R@1 ↑
R@5 ↑
R@10 ↑
AP ↑
0 (Initial BEV)
58.8
78.7
84.2
63.2
1
66.8
84.0
87.9
70.6
2
79.0
88.0
89.3
81.1
3
80.5
89.2
89.2
82.2
Table 3: Retrieval Results of BEV View Refinement.
Figure 8: Curve of Meter-level Position Success Rate. Meter-level Success Rate (MSR@X) is the percentage of queries localized within the X-meter threshold.
View Num.
E^RMSE↓
E^median↓
Δθ^RMSE↓
Δθ^median↓
4
4.36
2.71
1.84
1.24
12
4.65
2.85
1.96
1.03
18
6.93
2.91
3.37
1.07
Table 4: Result of Relative Pose Evaluation. We report the 3-DoF Relative Pose Error (RPE) of UAV frames.
(a)
(b)
(c)
R@1 ↑
Emedian↓
Δθmean↓
Δθmedian↓
67.2
60.1
22.9
5.2
✓
67.7
57.1
17.3
4.3
✓
71.6
42.1
19.3
4.2
✓
✓
72.3
37.8
20.7
4.8
✓
✓
79.0
26.2
12.2
4.1
Table 5: Ablation Study of the Proposed Modules. (a) Satellite-wise Attention (b) Geometric Validation (c) Feature Validation .
View Number
R@1 ↑
R@5 ↑
R@10 ↑
AP ↑
4
64.9
79.0
82.9
68.2
12
79.0
88.0
89.3
81.1
18
77.8
86.4
90.6
83.1
Table 6: Impact of the Number of Input Images.
Figure 9: Overlapped Satellite Images Distribution Comparison.
Cross-view geo-localization between UAV and satellite imagery remains a fundamental yet highly challenging task, especially under large off-nadir views where drastic perspective distortions, occlusions, and appearance gaps occur. Existing benchmarks and methods primarily focus on near-nadir scenarios and often overlook the importance of structural scene understanding and intra-domain relational constraints, limiting their performance in real-world deployments. In this work, we introduce OffNadirLoc, a new benchmark for large off-nadir UAV-to-satellite geo-localization. To tackle the unique challenges posed by off-nadir perspectives, we further propose ONLoc, a framework that incorporates a structure-aware contextual weighting mechanism to dynamically emphasize reliable local features while suppressing ambiguous or repetitive regions. Additionally, we design a view-coherent learning strategy, which treats one satellite image and the corresponding UAV images from multiple views as a cohesive semantic group. This set-level supervision enables the model to learn viewpoint-invariant and discriminative features, making it more effective at capturing multi-view consistency than conventional pairwise contrastive learning. Extensive experiments on the OffNadirLoc benchmark and four near-nadir datasets demonstrate that our method consistently outperforms state-of-the-art approaches while exhibiting strong zero-shot generalization to unseen datasets without additional training. The code will be released at https://montalario.github.io/offnadirloc/.
Qian Qiao, Wenye Liu, Ting Liu +2
School of Computer Science, Northwestern Polytechnical University
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.
Kejia Liu, Haoyang Zhou, Ruoyu Xu +3
College of Computer Science and Technology, Zhejiang University · State Key Laboratory of Blockchain and Data Security, Zhejiang University · Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security
Cross-View Geo-Localization (CVGL) between UAV imagery and satellite images plays a crucial role in target localization and UAV self-positioning. However, most existing methods rely on the idealized assumption of scale consistency between UAV queries and satellite galleries, overlooking the severe scale ambiguity commonly encountered in real-world scenarios. This discrepancy leads to field-of-view misalignment and feature mismatch, significantly degrading CVGL robustness. To address this issue, we propose a geometric framework that recovers the absolute metric scale from monocular UAV images using semantic anchors. Specifically, small vehicles (SVs), characterized by relatively stable prior size distributions and high detectability, are exploited as metric references. A Decoupled Stereoscopic Projection Model is introduced to estimate the absolute image scale from these semantic targets. By decomposing vehicle dimensions into radial and tangential components, the model compensates for perspective distortions in 2D detections of 3D vehicles, enabling more accurate scale estimation. To further reduce intra-class size variation and detection noise, a dual-dimension fusion strategy with Interquartile Range (IQR)-based robust aggregation is employed. The estimated global scale is then used as a physical constraint for scale-adaptive satellite image cropping, improving UAV-to-satellite feature alignment. Experiments on augmented DenseUAV and UAV-VisLoc datasets demonstrate that the proposed method significantly improves CVGL robustness under unknown UAV image scales. Additionally, the framework shows strong potential for downstream applications such as passive UAV altitude estimation and 3D model scale recovery.
Yibin Ye, Shuo Chen, Kun Wang +5
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China · Hunan Provincial Key Laboratory of Image Measurement and Vision Navigation, Changsha 410073, China · School of Information Science and Engineering, Lanzhou University, Lanzhou 730000, China