Training-free attentive-patch selection for visual place recognition

Visual Place Recognition (VPR) utilizing patch descriptors from Convolutional Neural Networks (CNNs) has shown impressive performance in recent years. Existing works either perform exhaustive matching of all patch descriptors, or employ complex networks to select good candidate patches for further g...

Full description

Saved in:
Bibliographic Details
Main Authors: Zhang, Dongshuo, Wu, Meiqing, Lam, Siew-Kei
Other Authors: College of Computing and Data Science
Format: Conference or Workshop Item
Language:English
Published: 2024
Subjects:
Online Access:https://hdl.handle.net/10356/178533
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
id sg-ntu-dr.10356-178533
record_format dspace
spelling sg-ntu-dr.10356-1785332024-06-25T08:54:14Z Training-free attentive-patch selection for visual place recognition Zhang, Dongshuo Wu, Meiqing Lam, Siew-Kei College of Computing and Data Science School of Computer Science and Engineering 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) Hardware & Embedded Systems Lab (HESL) Computer and Information Science Robotics Simultaneous localization and mapping Visual place recognition Visual Place Recognition (VPR) utilizing patch descriptors from Convolutional Neural Networks (CNNs) has shown impressive performance in recent years. Existing works either perform exhaustive matching of all patch descriptors, or employ complex networks to select good candidate patches for further geometric verification. In this work, we develop a novel two-step training-free patch selection method that is fast, while being robust to large occlusions and extreme viewpoint variations. In the first step, a self-attention mechanism is used to select sparse and evenly distributed discriminative patches in the query image. Next, a novel spatial-matching method is used to rapidly select corresponding patches with high similar appearances between the query and each reference image. The proposed method is inspired by how humans perform place recognition by first identifying prominent regions in the query image, and then relying on back-and-forth visual inspection of the query and reference image to attentively identify similar regions while ignoring dissimilar ones. Extensive experiment results show that our proposed method outperforms state-of-the-art (SOTA) methods in both place recognition precision and runtime, on various challenging conditions. Ministry of Education (MOE) Submitted/Accepted version This work was supported in part by the Ministry of Education, Singapore, under its IEO Decentralized Funding, under Grant NGF-2020-09-028; and in part by the Ministry of Education, Singapore, under its Academic Research Fund Tier 1, under Grant RG78/21. 2024-06-25T08:54:14Z 2024-06-25T08:54:14Z 2023 Conference Paper Zhang, D., Wu, M. & Lam, S. (2023). Training-free attentive-patch selection for visual place recognition. 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 9169-9174. https://dx.doi.org/10.1109/IROS55552.2023.10342347 978-1-6654-9190-7 https://hdl.handle.net/10356/178533 10.1109/IROS55552.2023.10342347 9169 9174 en NGF-2020-09-028 RG78/21 © 2023 IEEE. All rights reserved. This article may be downloaded for personal use only. Any other use requires prior permission of the copyright holder. The Version of Record is available online at http://doi.org/10.1109/IROS55552.2023.10342347. application/pdf
institution Nanyang Technological University
building NTU Library
continent Asia
country Singapore
Singapore
content_provider NTU Library
collection DR-NTU
language English
topic Computer and Information Science
Robotics
Simultaneous localization and mapping
Visual place recognition
spellingShingle Computer and Information Science
Robotics
Simultaneous localization and mapping
Visual place recognition
Zhang, Dongshuo
Wu, Meiqing
Lam, Siew-Kei
Training-free attentive-patch selection for visual place recognition
description Visual Place Recognition (VPR) utilizing patch descriptors from Convolutional Neural Networks (CNNs) has shown impressive performance in recent years. Existing works either perform exhaustive matching of all patch descriptors, or employ complex networks to select good candidate patches for further geometric verification. In this work, we develop a novel two-step training-free patch selection method that is fast, while being robust to large occlusions and extreme viewpoint variations. In the first step, a self-attention mechanism is used to select sparse and evenly distributed discriminative patches in the query image. Next, a novel spatial-matching method is used to rapidly select corresponding patches with high similar appearances between the query and each reference image. The proposed method is inspired by how humans perform place recognition by first identifying prominent regions in the query image, and then relying on back-and-forth visual inspection of the query and reference image to attentively identify similar regions while ignoring dissimilar ones. Extensive experiment results show that our proposed method outperforms state-of-the-art (SOTA) methods in both place recognition precision and runtime, on various challenging conditions.
author2 College of Computing and Data Science
author_facet College of Computing and Data Science
Zhang, Dongshuo
Wu, Meiqing
Lam, Siew-Kei
format Conference or Workshop Item
author Zhang, Dongshuo
Wu, Meiqing
Lam, Siew-Kei
author_sort Zhang, Dongshuo
title Training-free attentive-patch selection for visual place recognition
title_short Training-free attentive-patch selection for visual place recognition
title_full Training-free attentive-patch selection for visual place recognition
title_fullStr Training-free attentive-patch selection for visual place recognition
title_full_unstemmed Training-free attentive-patch selection for visual place recognition
title_sort training-free attentive-patch selection for visual place recognition
publishDate 2024
url https://hdl.handle.net/10356/178533
_version_ 1806059850747084800