Graph guided method improves alignment of large aerial image collections
G3AR: Graph-Guided Neural Visual Geometry for Scalable Multi-Sequence Aerial Registration
Computer Vision and Pattern Recognition
Summary
Aligning thousands of aerial images to build accurate 3D maps is very hard because existing methods either can’t handle many images at once or don’t properly handle tricky overlaps between image groups. The authors propose a new method called G3AR that uses a graph to guide how images are grouped and aligned, improving accuracy and speed. They tested G3AR on real aerial scenes and found it works better than other neural methods for this task. This method helps stitch together big aerial photo collections into precise maps.
What this means in practice
- •For geospatial mapping teams: Improve accuracy and speed of aligning thousands of aerial images for large-scale 3D map creation.
- •For drone operation teams: Register aerial image sequences from drones into a common frame to support precise site surveying and monitoring.
Authors
Jeng Wen Joshua Lean, Ting-Yu Yen, Wei-Fang Sun, Simon See, Hung-Kuo Chu, Shih-Hsuan Hung
Abstract
Full-context neural visual geometry is impractical for thousands of images, while sequence-based chunking poorly captures irregular non-local overlap in multi-sequence aerial collections. We present Graph-Guided Neural Visual Geometry for Aerial Registration (G3AR), a graph-guided framework for scalable dense neural geometry. Before local inference, G3AR builds a geometrically verified image-proximity graph that guides bounded overlapping chunks and induces a chunk graph whose maximum spanning tree defines alignment topology. Compatible backbones process chunks independently; shared-image predictions then estimate three-dimensional similarity (Sim(3)) transforms that register local cameras and geometry in a common frame. Across four real aerial scenes, G3AR improves pose error and runtime in matched VGGT- and Pi3-backed comparisons, while its DA3 variant achieves the lowest pose error among evaluated neural-geometry methods.