Exploring the best way for UAV visual localization under Low-altitude Multi-view Observation Condition: a Benchmark

TL;DR

The paper tackles absolute UAV visual localization (AVL) under challenging low-altitude multi-view conditions using 2.5D reference maps. It introduces AnyVisLoc, a large-scale dataset, and a unified framework that combines image retrieval, pixel-level image matching, and DSM-assisted PnP localization. Through extensive experiments, it benchmarks state-of-the-art AVL methods, introduces the PDM@K retrieval metric, and identifies CAMP+Roma+Top-N Re-rank as a strong baseline achieving 74.1% localization within 5 m. The work highlights how reference-map type, pitch angle, and prior-information noise influence performance and provides a public dataset and code to drive future advancements in low-altitude multi-view UAV AVL.

Abstract

Absolute Visual Localization (AVL) enables Unmanned Aerial Vehicle (UAV) to determine its position in GNSS-denied environments by establishing geometric relationships between UAV images and geo-tagged reference maps. While many previous works have achieved AVL with image retrieval and matching techniques, research in low-altitude multi-view scenarios still remains limited. Low-altitude Multi-view condition presents greater challenges due to extreme viewpoint changes. To explore the best UAV AVL approach in such condition, we proposed this benchmark. Firstly, a large-scale Low-altitude Multi-view dataset called AnyVisLoc was constructed. This dataset includes 18,000 images captured at multiple scenes and altitudes, along with 2.5D reference maps containing aerial photogrammetry maps and historical satellite maps. Secondly, a unified framework was proposed to integrate the state-of-the-art AVL approaches and comprehensively test their performance. The best combined method was chosen as the baseline and the key factors that influencing localization accuracy are thoroughly analyzed based on it. This baseline achieved a 74.1% localization accuracy within 5m under Low-altitude, Multi-view conditions. In addition, a novel retrieval metric called PDM@K was introduced to better align with the characteristics of the UAV AVL task. Overall, this benchmark revealed the challenges of Low-altitude, Multi-view UAV AVL and provided valuable guidance for future research. The dataset and codes are available at https://github.com/UAV-AVL/Benchmark

Figures (7)

• Figure 1: Benchmark Overview. This benchmark focuses on UAV visual localization under low-altitude multi-view observation condition using the 2.5D aerial or satellite reference maps. The visual localization is mainly achieved via a unified framework combining image retrieval, image matching, and PnP problem solving.
• Figure 2: Pitch Angle and Flight Altitude Distribution.
• Figure 3: Dataset Overview. The AnyVisLoc dataset contians Multi-scene, Multi-altitude, and Multi-view UAV images taken in 15 cities across China, as well as aerial and satellite reference maps. Each UAV image shows its flight altitude and pitch angle below.
• Figure 4: Illustration of PDM@K. (a) Different retrieval metric comparison. For clarity in the same figure, we have converted the spatial distance $d_i$ of SDM@1 to $R_i$ and the threshold for Recall@1 is set to 0.5. (b) Different parameter combinations for PDM@1. (c) Relation between localization accuracy and $R_i$. This curve is based on the actual AVL experiment results.
• Figure 5: Visualization of the relation between $R_i$ in PDM@K and subsequent localization accuracy.