BROWSE
Detail
Journal Article
TileLoc: Sequential Visual Place Recognition of UAVs using Public Tile Maps
Cited - time in 
Share 
Share
- Authors
- Dongwook Heo, Seungpyo Jeon, Jinwon Choi, Dayena Jeong, Hyeonjoong Wi, In Sung Jang, Sunglok Choi
- Issue Date
- 2026-09
- Citation
- IEEE Robotics and Automation Letters, v.11, no.9, pp.10601-10608
- ISSN
- 2377-3766
- Publisher
- IEEE
- Language
- English
- Type
- Journal Article
- Abstract
- We propose TileLoc, a visual place recognition method for uncrewed aerial vehicles (UAVs) that uses publicly available web tile maps in GNSS-degraded or GNSS-denied environments without prior mapping. Visual place recognition with web tile maps is challenging because the query image and the map tiles often differ in scale, rotation, and coverage. As a result, single-frame retrieval can be unreliable. Based on our preliminary experiments, we identified major failure cases and designed a simple yet effective pipeline to mitigate them. TileLoc builds a multi-zoom, overlapped tile database, uses a set of rotated query views, and stabilizes retrieval results by accumulating evidence over time using log-odds voting with temporal decay. After the initial global search, TileLoc switches to local search when its confidence becomes high enough, reducing confusion caused by visually similar tiles in global search. Experimental results showed consistent recall improvements over a strong global retrieval baseline, achieving up to 84.5% Recall@5 within a 250 m threshold. These results demonstrate the effectiveness of TileLoc for UAV visual place recognition using public tile maps in GNSS-degraded or GNSS-denied environments.
- Keyword
- sequential matching, UAV, Visual place recognition, web tile map
- KSP Keywords
- Global Search, Local search, Map Tile, Over time, Recognition method, Visual place recognition, aerial vehicle, query image, temporal decay