Beyond Paired Data: Self-Supervised UAV Geo-Localization from Reference Imagery Alone

2025-12-02

AI Summary

To address the limitations of existing cross-view geo-localization methods for UAVs in GNSS-denied environments—namely, their reliance on large-scale paired UAV-satellite image datasets, high data acquisition costs, and poor generalizability—this paper proposes the first self-supervised, satellite-only cross-view localization framework. Our method models the visual domain shift between satellite and UAV viewpoints without requiring paired samples, enabling a contrastive self-supervised learning paradigm. We introduce CAEVL, a lightweight contrastive autoencoder architecture, coupled with viewpoint-aware data augmentation strategies tailored to inter-modal geometric and appearance discrepancies. Evaluated on the newly released ViLD real-world UAV dataset, our approach achieves localization accuracy comparable to fully supervised paired-training baselines—even under zero-shot pairing—while significantly enhancing generalization in low-resource settings. This work establishes an efficient, scalable, and GNSS-free visual localization paradigm.

Technology Category

• Computer Vision: Multi-modal Vision
• Intelligent Robots: Localization, Mapping, and Navigation
• Machine Learning: Multi-instance/Multi-view Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Image-based localization in GNSS-denied environments is critical for UAV autonomy. Existing state-of-the-art approaches rely on matching UAV images to geo-referenced satellite images; however, they typically require large-scale, paired UAV-satellite datasets for training. Such data are costly to acquire and often unavailable, limiting their applicability. To address this challenge, we adopt a training paradigm that removes the need for UAV imagery during training by learning directly from satellite-view reference images. This is achieved through a dedicated augmentation strategy that simulates the visual domain shift between satellite and real-world UAV views. We introduce CAEVL, an efficient model designed to exploit this paradigm, and validate it on ViLD, a new and challenging dataset of real-world UAV images that we release to the community. Our method achieves competitive performance compared to approaches trained with paired data, demonstrating its effectiveness and strong generalization capabilities.

Problem

• Enables UAV localization without GNSS using only satellite images
• Eliminates need for paired UAV-satellite datasets during training
• Simulates visual domain shift between satellite and UAV views

Innovation

• Self-supervised learning from satellite images alone
• Augmentation simulates satellite-to-UAV domain shift
• Efficient CAEVL model validated on real-world UAV dataset