University of Twente Student Theses

Login

Visual place recognition under image corruptions

Smit, P.J.M. (2022) Visual place recognition under image corruptions.

[img] PDF
11MB
Abstract:The field of Visual Place Recognition (VPR) is concerned with finding out where an input/query photo was taken by retrieving similar geotagged images. Not all queries are perfect; some may contain corruptions like motion blur, compression, bit errors, etc. The goal of this research is to investigate the robustness of current VPR strategies against such corruptions and give insights into how it can be improved in future works. We thoroughly evaluated the robustness by introducing novel metrics. Out of the three network architectures investigated, ResNeXt-101 32x8d performs the best. Also, we found that a GeM pooling layer and Generalized Contrastive Loss function often improve corruption robustness over traditional methods. We also give some insights on evaluating long-term corruptions in VPR.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:https://purl.utwente.nl/essays/89453
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page