A Hierarchical Dual Model of Environment- and Place-Specific Utility for Visual Place Recognition

Visual Place Recognition (VPR) approaches have typically attempted to match places by identifying visual cues, image regions or landmarks that high “utility” in a specific place. But this concept of utility is not singular - rather it can take range forms. In letter, we present novel approach deduce two key types for VPR: the cues `specific' an environment, and particular We employ contrastive learning principles estimate both environment- place-specific Vector Locally Aggregated Descriptors (VLAD) clusters unsupervised manner, which then used guide local feature matching through keypoint selection. By combining these measures, our achieves state-of-the-art performance on three challenging benchmark datasets, while simultaneously reducing required storage compute time. provide further analysis demonstrating cluster selection results semantically meaningful results, finer grained categorization often has higher VPR than level semantic (e.g. building, road), characterise how measures vary across different environments. Source code made publicly available at https://github.com/Nik-V9/HEAPUtil.