Abstract:In this paper, we address the problem of landmark-based visual place recognition. In the state-of-the-art method, accurate object proposal algorithms are first leveraged for generating a set of local regions containing particular landmarks with high confidence. Then, these candidate regions are represented by deep features and pairwise matching is performed in an exhaustive manner for the similarity measure. Despite its success, conventional object proposal methods usually produce massive landmark-dependent image patches exhibiting significant distribution variance in scale and overlap. As a result, the inconsistency in landmark distributions tends to produce biased similarity between pairwise images yielding the suboptimal performance. In order to gain an insight into the landmark-based place recognition scheme, we conduct a comprehensive study in which the influence of landmark scales and the proportion of overlap on the recognition performance is explored. More specifically, we thoroughly study the exhaustive search based landmark matching mechanism, and thus derive three-fold important observations in terms of the beneficial effect of specific landmark generation strategies. Inspired by the above observations, a simple yet effective dense sampling based scheme is presented for accurate place recognition in this paper. Different from the conventional object proposal strategy, we generate local landmarks of multiple scales with uniform distribution from entire image by dense sampling, and subsequently perform multi-scale fusion on the densely sampled landmarks for similarity measure. The experimental results on three challenging datasets demonstrate that the recognition performance can be significantly improved by our efficient method in which the landmarks are appropriately produced for accurate pairwise matching.

Place Recognition and Online Learning in Dynamic Scenes with Spatio-Temporal Landmarks.

Generative Methods for Long-Term Place Recognition in Dynamic Scenes

A Robust Image-Sequence-Based Framework for Visual Place Recognition in Changing Environments.

Dynamic Scene Models For Incremental, Long-Term, Appearance-Based Localisation

Unsupervised Learning Methods for Visual Place Recognition in Discretely and Continuously Changing Environments

Structure-Aware Feature Disentanglement with Knowledge Transfer for Appearance-Changing Place Recognition

Multi-Modal Visual Place Recognition in Dynamics-Invariant Perception Space.

Discriminative and Semantic Feature Selection for Place Recognition Towards Dynamic Environments

An Efficient Point Cloud Place Recognition Approach Based on Transformer in Dynamic Environment

LiDAR-based Semantic Place Recognition in Dynamic Urban Environments

SC-LPR: Spatiotemporal Context Based LiDAR Place Recognition

Improving Place Recognition Using Dynamic Object Detection

Towards A Deep Insight into Landmark-based Visual Place Recognition: Methodology and Practice.

BEVPlace: Learning LiDAR-based Place Recognition using Bird's Eye View Images

STV-SC: Segmentation and Temporal Verification Enhanced Scan Context for Place Recognition in Unstructured Environment

Season-Invariant and Viewpoint-Tolerant LiDAR Place Recognition in GPS-Denied Environments

Learning Sequence Descriptor based on Spatio-Temporal Attention for Visual Place Recognition

Stabilize an Unsupervised Feature Learning for LiDAR-based Place Recognition

Leveraging variable sensor spatial acuity with a homogeneous, multi-scale place recognition framework

Localizing Discriminative Visual Landmarks for Place Recognition

Learning robust representation and sequence constraint for retrieval-based long-term visual place recognition