PiLoT v2: Pixel-to-Orthogonal Map Alignment for Free-view UAV Geo-localization

2026-06-29

AI Summary

This work addresses the high storage and computational overhead incurred by heavy 3D maps in real-time, drift-free drone localization under GNSS-denied environments. To this end, we propose a lightweight localization method that replaces conventional 3D maps with true digital orthophoto maps (TDOM) and digital surface models (DSM), substitutes GPU-intensive rendering with CPU-friendly map cropping, and introduces a novel direct alignment paradigm from image pixels to orthorectified maps. By leveraging a large-scale geometrically annotated dataset to train a cross-view feature matching network and incorporating gravity direction and single-point LiDAR ranging as priors to refine the pose manifold, our approach achieves localization accuracy comparable to or surpassing existing pixel-to-3D methods while substantially reducing resource demands, thereby enabling efficient deployment on embedded platforms.

Abstract

Real-time, drift-free UAV geo-localization is essential for autonomous missions in GNSS-denied environments. The pioneering system, PiLoT, achieves high precision via Neural Pixel-to-3D Registration, aligning UAV video streams with a single rendered reference view from 3D meshes. However, its reliance on heavy 3D meshes incurs massive storage overheads, complex map acquisition, and significant computational rendering costs, severely hindering deployment on embedded platforms. To address these bottlenecks, we propose PiLoT v2, a lightweight yet robust evolution that shifts the paradigm to direct pixel-to-orthogonal map registration for free-view UAV geo-localization. By leveraging True Digital Orthophoto Maps (TDOMs) and Digital Surface Models (DSMs) as the reference substrate, PiLoT v2 replaces GPU-intensive 3D rendering with a highly efficient, CPU-friendly map cropping operation. To bridge the severe geometric discrepancy between these 2.5D orthogonal crops and free-view oblique UAV imagery, we train a cross-view feature registration network using a novel, large-scale geometrically annotated dataset. Furthermore, we integrate onboard sensor prior--specifically gravity direction and single-point laser rang--directly into the pose optimization manifold to enhance robustness against cross-view visual degradation. Experimental results demonstrate that PiLoT v2 achieves performance comparable to, or even exceeding, its Pixel-to-3D predecessor, while offering drastically lower storage and computational costs.

Problem

• UAV geo-localization
• GNSS-denied
• 3D mesh
• storage overhead
• computational cost

Innovation

• orthogonal map registration
• cross-view feature matching
• lightweight geo-localization
• sensor-fused pose optimization
• UAV localization