Altitude-Adaptive Vision-Only Geo-Localization for UAVs in GPS-Denied Environments

Xingyu Shao; Mengfan He; Liangzheng Sun; Chunyu Li; Ziyang Meng

cs.CVFeb 27, 2026

Altitude-Adaptive Vision-Only Geo-Localization for UAVs in GPS-Denied Environments

Authors: Xingyu Shao, Mengfan He, Liangzheng Sun, Chunyu Li, Ziyang Meng

Organizations: Department of Precision Instrument, Tsinghua University, Beijing 100084, China · School of Instrumentation Science and Optoelectronics Engineering, Beijing Information Science and Technology University, Beijing 100192, China · School of Aerospace Engineering, Beijing Institute of Technology, Beijing 100081, China

Abstract

Matching downward-looking unmanned aerial vehicle (UAV) images to georeferenced satellite or aerial map tiles supports local earth observation, map-based interpretation, and coarse geo-initialization when direct positioning signals are degraded or unavailable. A major difficulty in this cross-platform matching task is the scale mismatch caused by large variations in UAV altitude. To address this problem, we propose an altitude-adaptive aerial visual place recognition framework. The method first estimates relative altitude from a single downward-looking image by transforming the input into the frequency domain and formulating altitude estimation as a regression-as-classification problem. The estimated altitude is then used to crop the query image to a canonical scale, after which a classification-then-retrieval visual place recognition module performs coarse map localization. To improve retrieval robustness under varying image quality, we further introduce a quality-adaptive margin classifier and refine the final location by weighted coordinate estimation over the top retrieved candidates. Experiments on two synthetic datasets and two real-flight datasets show that the relative altitude estimation module improves downstream retrieval performance under substantial altitude changes. With our visual place recognition module, altitude adaptation improves average R@1 and R@5 by 41.50 and 56.83 percentage points, respectively, compared with using the same retrieval pipeline without altitude normalization, and the main retrieval pipeline runs at 13.3 frames/s on the workstation used for evaluation. These results indicate that relative altitude estimation provides a useful scale prior for cross-altitude UAV-to-map image matching using georeferenced remote sensing imagery.

Figures & tables

Fig. 1: The overview of the proposed method: altitude variation causes scale mismatch, while frequency-domain altitude estimation and altitude-aware cropping normalize the query image to a canonical scale for more reliable cross-altitude VPR.

Fig. 2: Overview of the proposed altitude-adaptive geo-localization framework. The framework consists of two stages: offline preparation and online inference. During offline preparation, satellite map tiles cropped at multiple altitudes are transformed into frequency-domain samples to train the relative altitude estimation (RAE) model, while tiles cropped at the canonical altitude $H_{\text{db}}$ are used to train the visual place recognition (VPR) model and prepare the reference database. During online inference, a query image is first transformed into the frequency domain and fed into the stored RAE model, which predicts a relative-altitude bin through regression-as-classification (RAC). The estimated altitude is then used to crop the query to the canonical scale, producing a normalized query for the stored VPR model. The VPR model retrieves the top candidate references. A weighted coordinate estimation module can optionally be applied after retrieval when coordinate refinement is required; this module is not part of the main recall pipeline and is evaluated separately in Section IV-E5 .

Fig. 3: Examples of raw query images, frequency-domain representations, and primitive images from the four datasets. For visual clarity, the displayed frequency-domain images are brightness-adjusted.

Fig. 4: Transformation from an input image to the primitive image at the canonical altitude.

Fig. 5: Classification-then-retrieval pipeline for the primitive query image. The primitive query image $I_{\text{q}}$ is encoded by an EfficientNet-B5 backbone and a MixVPR aggregator. Group-wise classifiers trained with the quality-adaptive margin classifier (QAMC) produce class scores over the spatial groups, from which the top $n_{\text{class}}$ predicted classes $\mathcal{C}_{\text{top}}$ are selected. Their corresponding sub-databases are merged into the candidate retrieval database $\mathrm{DB}_{\text{retrieval}}$ , where feature retrieval returns the top $n_{\text{retrieve}}$ reference descriptors and their associated UTM coordinates.


Sensor	Details

Camera	Type : FLIR BFS-U3-31S4C-C RGB channels, 2048 $\times$ 1536 resolution, 55Hz max frame rate (20Hz in experiment), global shutter
Lens	Type : Chiopt FA0401C 82.9 ∘ horizontal FOV, 66.5 ∘ vertical FOV, 4 $\sim$ 75 mm focal length
GNSS	Type : NovAtel OEM718D Dual-antenna, 5 Hz update, 1.5 m (RMS) with single point, 1cm + 1ppm (RMS) with RTK
Altimeter	Type : Benewake TF350 10Hz update, 350m maximum detection range with 0.1m accuracy

TABLE II: Specifications of the UAV-mounted sensing and imaging hardware


Metric	RAE	Crop	VPR (C)	VPR (R)

Avg latency (ms)	10.7	12.4	50.6	1.5
Max latency (ms)	14.0	36.5	92.7	4.0
Peak CPU ABS (MB)	260.57	679.81	1830.71	1830.74
CPU DELTA (MB)	0.48	21.09	70.45	2.74

TABLE V: Runtime statistics of the proposed pipeline


$\mathrm{WCE}$	CT01	CT02	QD01	QD02

✓	96.20	87.00	69.78	59.57
✗	77.20	68.40	54.67	47.45

TABLE X: Localization success rate

S_{\text{Loc}}

with and without weighted coordinate estimation (WCE)

Cross-View Geo-Localization

Cross-view geo-localization matches ground, drone, or aerial images to maps or satellite views to estimate where they were captured.

See all 94 papers

RIM: A Retrieval-In-Matching Framework for Cross-Domain Global Visual Localization of UAVs
Jul 22, 2026Xin Li, Siyuan Duan, Shang Wang +3Cross-View Geo-Localization Visual Place Recognition
Key Laboratory of Spectral Imaging Technology CAS, Xi’an Institute of Optics and Precision Mechanics, Chinese Academy of Sciences, Xi’an, 710119, China · University of Chinese Academy of Sciences, Beijing, 100049, China
DECO: Depth-Guided Co-Visibility Reasoning for Low-Altitude UAV Visual Localization
Aug 23, 2026Yibin Ye, Xichao Teng, Shuo Chen +4GNSS-Denied Localization Map-Based Localization
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China
NGPS: GPS-Denied Aerial Geo-Localization and 2.5D Reconstruction via Deep Satellite Image Matching and Multi-Rate Sensor Fusion
Jul 21, 2026Sanket SharmaCross-View Geo-Localization UAV Navigation
Independent Researcher

Jul 22, 2026J/K move · Enter open · S save

RIM: A Retrieval-In-Matching Framework for Cross-Domain Global Visual Localization of UAVs

Xin Li, Siyuan Duan, Shang Wang, Zhimin Mao+2

Key Laboratory of Spectral Imaging Technology CAS, Xi’an Institute of Optics and Precision Mechanics, Chinese Academy of Sciences, Xi’an, 710119, China · University of Chinese Academy of Sciences, Beijing, 100049, China

Cross-View Geo-Localization Visual Place Recognition UAV Navigation Local Feature Matching

Open paper AskPDF arXiv

Global visual localization of unmanned aerial vehicles (UAVs) using remote-sensing reference maps has attracted increasing attention. However, acquisition-time and imaging-platform differences between UAV and reference imagery induce substantial cross-domain appearance and viewpoint shifts, challenging robust six-degree-of-freedom (6-DoF) pose estimation. We address these shifts by sampling UAV-viewpoint reference views from Google 3D Tiles across locations, altitudes, and orientations. A two-stage cross-domain fine-tuning recipe adapts SALAD using pose-near positives and geographically distant hard negatives, while local geometric consistency re-ranks the Top-K candidates. We further propose Retrieval-In-Matching (RIM), which freezes the adapted DINOv2-B retriever and distils a local-descriptor decoder that reuses its token field alongside a shallow VGG19 detail stream. One query-side DINOv2-B forward thus serves both SALAD retrieval and local description, eliminating a second foundation-model backbone while preserving retrieval descriptors by construction. We evaluate RIM zero-shot on the reconstructed EPFL Urbanscape and self-collected Chang'an Park datasets, both geographically disjoint from the training data. RIM outperforms ten recent retrieval baseline families. At 25/50 m under the full 3D distance metric, it improves Recall@1 over SALAD by 8.55/13.77 percentage points on EPFL and 4.45/8.94 points on Park. At Top-K=5, the complete measured localization query, including retrieval, candidate matching, and robust geometric verification, takes 67.9 ms end-to-end: 1.8 times faster than the strongest separate sparse-matching baseline and over 40 times faster than RoMa, while achieving comparable re-ranking accuracy. These results establish an efficient and deployable pipeline for UAV global visual localization in GNSS-challenged environments.

Abstract

Cross-View Geo-Localization

Explore similar work

RIM: A Retrieval-In-Matching Framework for Cross-Domain Global Visual Localization of UAVs

DECO: Depth-Guided Co-Visibility Reasoning for Low-Altitude UAV Visual Localization

NGPS: GPS-Denied Aerial Geo-Localization and 2.5D Reconstruction via Deep Satellite Image Matching and Multi-Rate Sensor Fusion