Exploring the best way for UAV visual localization under Low-altitude Multi-view Observation Condition: a Benchmark (2503.10692v1)

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

• The paper presents the AnyVisLoc dataset and a unified framework to benchmark UAV absolute visual localization (AVL) under challenging low-altitude, multi-view conditions, achieving a 74.1% localization accuracy baseline within 5 meters.
• The study identifies ConvNeXt-based CAMP retrieval combined with Roma dense matching as the most effective approach for AVL in these complex low-altitude scenarios.
• It highlights that localization performance is significantly impacted by factors such as multi-view observations (small pitch angles), reference map resolution, and noise in prior altitude/angle information.

UAV Visual Localization Under Low-Altitude Multi-View Observation: An Analytical Overview

The paper "Exploring the best way for UAV visual localization under Low-altitude Multi-view Observation Condition: a Benchmark" presents an in-depth paper of Absolute Visual Localization (AVL) for Unmanned Aerial Vehicles (UAVs) under challenging low-altitude and multi-view scenarios. Historically, AVL has been contingent on nadir-view assumptions with relatively planar terrains; however, this paper broadens the scope by tackling extreme viewpoint variations inherent in low-altitude, multi-view conditions.

Dataset Creation and Baseline Performance

The researchers developed the AnyVisLoc dataset, a comprehensive resource comprising 18,000 UAV images captured across varied terrains and altitudes, coupled meticulously with 2.5D reference maps. These maps integrate both aerial and satellite data, enabling thorough evaluation across existing AVL techniques. The rigorous dataset construction ensures considerable variability in viewpoint, pitch angles, and observation conditions, essential for benchmarking AVL methodologies in these challenging scenarios.

Central to their analysis, the researchers proposed a unified framework for evaluating AVL approaches. This framework integrates contemporary retrieval and matching algorithms, culminating in a baseline evaluation. A key achievement of their baseline methodology was obtaining a 74.1% localization accuracy within a 5-meter radius, recognized as a significant benchmark under such complex conditions.

Evaluation of Image Retrieval and Matching Techniques

The paper meticulously evaluates various image retrieval and matching methods, concluding that the ConvNeXt-based CAMP retrieval method, when combined with the Roma dense matching algorithm, provides the most effective localization results. This choice not only ensures a high retrieval accuracy—as denoted by a substantial Recall@K and the newly proposed PDM@K metrics—but also balances the intricacy of image viewpoint shifts, which is pivotal for tackling the inherited complexities of low-altitude scenarios.

Implications of Multi-View and Reference Maps

A critical component discussed is the impact of multi-view observations, particularly smaller pitch angles associated with oblique imagery, which traditionally degrade the effectivity of retrieval and matching accuracy. Moreover, the paper delineates the disparity in effectiveness between various reference map sources, highlighting the degradation in localization performance when lower-resolution satellite maps are utilized instead of high-resolution aerial maps.

Factors Influencing Baseline Performance

The paper underscores the sensitivity of AVL to prior information in terms of pitch/yaw angles and altitude. Notably, noise in these parameters can notably affect the localization outcome, especially in dynamically changing low-altitude environments. The paper exemplifies this by quantifying performance shifts under varying degrees of noise interference.

Future Directions

The insights derived from this research have multifaceted implications for future AVL methodologies and UAV operational frameworks. Enhancements in real-time localization accuracy under extreme conditions remain an essential goal. Future research may focus on further refining learning-based retrieval techniques and integrating advanced sensory data to mitigate the persisting challenges highlighted through the benchmark.

In conclusion, this paper contributes significantly to the UAV AVL domain by establishing a robust benchmark foundation for low-altitude multi-view environments, paving the way for advanced research in resilient UAV operations in GNSS-denied regions. The work not only advances scholarly discourse but also promises practical applications across diverse sectors where UAVs operate under dynamic real-world conditions.