Visual place recognition for aerial imagery: A survey

TL;DR

This paper introduces a methodology tailored for evaluating VPR techniques specifically in the domain of aerial imagery, providing a comprehensive assessment of various methods and their performance and demonstrates the importance of selecting appropriate zoom and overlap levels when constructing map tiles to achieve maximum efficiency of VPR algorithms in the case of aerial imagery.

Abstract

Aerial imagery and its direct application to visual localization is an essential problem for many Robotics and Computer Vision tasks. While Global Navigation Satellite Systems (GNSS) are the standard default solution for solving the aerial localization problem, it is subject to a number of limitations, such as, signal instability or solution unreliability that make this option not so desirable. Consequently, visual geolocalization is emerging as a viable alternative. However, adapting Visual Place Recognition (VPR) task to aerial imagery presents significant challenges, including weather variations and repetitive patterns. Current VPR reviews largely neglect the specific context of aerial data. This paper introduces a methodology tailored for evaluating VPR techniques specifically in the domain of aerial imagery, providing a comprehensive assessment of various methods and their performance. However, we not only compare various VPR methods, but also demonstrate the importance of selecting appropriate zoom and overlap levels when constructing map tiles to achieve maximum efficiency of VPR algorithms in the case of aerial imagery. The code is available on our GitHub repository -- https://github.com/prime-slam/aero-vloc.

Figures (9)

• Figure 1: Left: sample frames from the used aerial datasets for Visual Place Recognition (VPR). Center: Recall@1 statistics for VPR methods without re-ranking step. Right: Recall@1 statistics for VPR methods with re-ranking step, top-100 candidates are used everywhere.
• Figure 2: Pipeline of the proposed aerial visual geo-localization system. The offline phase involves the computation of global descriptors and local features for map tiles. During the online phase, the VPR method selects the $N$ nearest images from the database, a re-ranking method identifies the optimal frame from the candidate set, and the precise location is determined through local alignment.
• Figure 3: Different zoom levels. Black means raw tiles, red means constructed tiles. Left: zoom 200%. Center: zoom 100%. Right: zoom 50%.
• Figure 4: Different overlap levels. The digits represent the numbering of the map tiles. Left: overlap 0%. Center: overlap 25%. Right: overlap 50%.
• Figure 5: A schematic representation of the Georeference Recall metric