University of Twente Student Theses

Login

Enhancing Robot Camera Localization through Image Retrieval via Generalized Contrastive Loss and Re-Ranking

Borkar, S.D. (2023) Enhancing Robot Camera Localization through Image Retrieval via Generalized Contrastive Loss and Re-Ranking.

Full text not available from this repository.

Full Text Status:Access to this publication is restricted
Abstract:The paper investigates the use of image retrieval for Visual Place Recognition (VPR) and its potential for visual localization. VPR is the task of recognizing a place based on its visual appearance, while visual localization involves estimating the precise location of a camera within that place. Existing methods for visual place recognition rely on handcrafted features or deep neural networks to extract discriminative information from images. Over the years, multiple methods have been proposed that are robust and obtain reliable results. In this paper, a state-of-the-art model using generalized contrastive loss (GCL) and graded similarity ground truth for image retrieval is examined and a re-ranking method is proposed to improve the VPR performance and visual localization. The re-ranking method is evaluated on the benchmark dataset - the RobotCar Seasons dataset v2 indicating that it outperforms state-of-the-art methods in visual localization. A comparative analysis of the one-stage retrieval and re-ranking methods is discussed. The competitive results suggest that image retrieval using the re-ranking method can be a powerful tool for visual place recognition and visual localization, with utility in applications like autonomous driving, and augmented reality that require high precision.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Electrical Engineering MSc (60353)
Link to this item:https://purl.utwente.nl/essays/94916
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page