TY - GEN

T1 - Dominating Set Database Selection for Visual Place Recognition

AU - Корнилова, Анастасия Валерьевна

AU - Москаленко, Иван Николаевич

AU - Пушкин, Тимофей Дмитриевич

AU - Tojiboev, Fakhriddin

AU - Tariverdizadeh, Rahim

AU - Ferrer, Gonzalo

N1 - Conference code: 21

PY - 2023/12/7

Y1 - 2023/12/7

N2 - This paper introduces a novel approach for creating a visual place recognition (VPR) database for localization in indoor environments from RGBD scanning sequences. The proposed method formulates the problem as a minimization challenge by utilizing a dominating set algorithm applied to a graph constructed from spatial information, referred to as the “DominatingSet” algorithm. Experimental results on various datasets, including 7-scenes, BundleFusion, RISEdb, and a specifically recorded sequences in a highly repetitive office setting, demonstrate that our technique significantly reduces database size while maintaining comparable VPR performance to state-of-the-art approaches in challenging environments. Additionally, our solution enables weakly-supervised labeling for all images from the sequences, facilitating the automatic fine-tuning of VPR algorithm to target environment. Additionally, this paper presents a fully automated pipeline for creating VPR databases from RGBD scanning sequences and introduces a set of metrics for evaluating the performance of VPR databases. The code and released data are available on our web-page — https://prime-slam.github.io/place-recognition-db/.

AB - This paper introduces a novel approach for creating a visual place recognition (VPR) database for localization in indoor environments from RGBD scanning sequences. The proposed method formulates the problem as a minimization challenge by utilizing a dominating set algorithm applied to a graph constructed from spatial information, referred to as the “DominatingSet” algorithm. Experimental results on various datasets, including 7-scenes, BundleFusion, RISEdb, and a specifically recorded sequences in a highly repetitive office setting, demonstrate that our technique significantly reduces database size while maintaining comparable VPR performance to state-of-the-art approaches in challenging environments. Additionally, our solution enables weakly-supervised labeling for all images from the sequences, facilitating the automatic fine-tuning of VPR algorithm to target environment. Additionally, this paper presents a fully automated pipeline for creating VPR databases from RGBD scanning sequences and introduces a set of metrics for evaluating the performance of VPR databases. The code and released data are available on our web-page — https://prime-slam.github.io/place-recognition-db/.

KW - Location awareness

KW - Visualization

KW - Codes

KW - Pipelines

KW - Indoor environment

KW - Visual databases

KW - Robots

U2 - 10.1109/ICAR58858.2023.10406721

DO - 10.1109/ICAR58858.2023.10406721

M3 - Conference contribution

SN - 979-8-3503-4230-7

T3 - International Conference on Advanced Robotics (ICAR)

SP - 473

EP - 479

BT - 2023 21st International Conference on Advanced Robotics (ICAR)

PB - Institute of Electrical and Electronics Engineers Inc.

CY - IEEE Xplore

T2 - 21st International Conference on Advanced Robotics (ICAR 2023)

Y2 - 5 December 2023 through 8 December 2023

ER -