During her PhD studies, Olaya Álvarez-Tuñón focused on enhancing underwater visual localization using deep learning. Cameras, while valuable, face challenges like floating particles and light dispersion underwater. Her research emphasized pose representation and loss function choices, examining their impact on network performance. She compared various pose regressor architectures and proposed new models better suited to underwater environments, improving the ability to learn robust features and generalize with limited data. To address data scarcity, she introduced a simulated dataset and a real dataset from pipeline inspections, demonstrating the potential for significant advancements in underwater navigation.

In summary, this work presents a comprehensive approach to enhancing underwater visual localization through deep learning, addressing data collection challenges, and optimizing network architectures to significantly advance navigation in complex underwater settings.

The PhD study was completed at the Department of Electrical and Computer Engineering, Faculty of Technical Sciences, Aarhus University.

This summary was prepared by the PhD student.

Time: Thursday, 6 September 2024 at 9:00
Place: Building 5122, room 122, Electrical and Computer Engineering, Aarhus University, Findlandsgade 20, 8200 Aarhus N
Title of PhD thesis: Vision-based navigation for underwater safety-critical applications
Contact information: Olaya Álvarez-Tuñón, e-mail: olaya.tunon@gmail.com, tel.: +45 55229905
Members of the assessment committee:
Guillermo Gallego, Professor at the Technische Universität Berlin, Germany.
Lazaros Nalpantidis, Professor of Autonomous Systems at the Department of Electrical and Photonics Engineering, Technical University of Denmark (DTU)
Rune Hylsberg Jacobsen (chair), Professor at the Department of Electrical and Computer Technology - Communication, Control and Automation, Aarhus University, Denmark
Main supervisor:
Henrik Karstoft Electrical and Computer Engineering - Signal Processing and Machine learning, Aarhus University, Denmark
Co-supervisor:
Erdal Kayacan, Full Professor, Automatic Control Group (RAT) Paderborn University, Germany
Yury Brodskiy, Senior Software Engineer, Computer Vision Group Leader, EIVA A/S, Denmark

Language: The PhD dissertation will be defended in English
The defence is public.
The PhD thesis is available for reading at the Graduate School of Technical Sciences/GSTS,

Ny Munkegade 120, build. 1521, 8000 Aarhus C.