Logo eprints

From Subpixel Accuracy to Scanpaths Analysis: Smart Strategies for Implementing Deep Learning Algorithms in Eye Movement Research and Applications

Byrne, Sean Anthony (2024) From Subpixel Accuracy to Scanpaths Analysis: Smart Strategies for Implementing Deep Learning Algorithms in Eye Movement Research and Applications. Advisor: Riccaboni, Prof. Massimo. Coadvisor: Polonio, Prof. Luca . pp. 181. [IMT PhD Thesis]

[img] Text (Doctoral thesis)
Thesis_Sean Byrne_2024_final (4).pdf - Published Version
Restricted to IMT staff and National library only until 30 April 2025.
Available under License Creative Commons Attribution Non-commercial Share Alike.

Download (10MB) | Request a copy


Eye-tracking research has been influential across various sec- tors, encompassing both the creation of eye-tracking devices and the analysis of the data they produce. These facets are known as gaze estimation and gaze analysis. The former identifies where an individual is gazing based on images cap- tured by cameras aimed at the eyes, while the latter discerns the duration and sites of gaze, typically using characteris- tics like saccades and fixations to deduce an individual’s cog- nitive activities. Recently, a significant transformation has taken place with both fields now heavily leaning on deep learning. This integration of deep learning methods has sig- nificantly improved precision, efficiency, and adaptability in both realms. It also ushers in advanced implementations, such as real-time gaze forecasting in areas like virtual real- ity and gaming. Yet, the infusion of deep learning comes with its set of challenges, notably when faced with the limited and often expensive eye-tracking datasets. This dissertation delves into these issues, focusing on the role of deep learning in both gaze estimation and analysis. Amongst the myriad of deep learning techniques for eye tracking, this work high- lights two: first, the efficacy of using synthetic data in gaze estimation models and its performance in synthetic and real- world pipelines. Second, within the context of an economic experiment, we investigate the impact of feature engineering for scanpath formulation and the potential to foresee a user’s choice before they decide, a concept that holds significance in numerous sectors, especially as eye tracking devices such as virtual headsets gain traction.

Item Type: IMT PhD Thesis
Subjects: R Medicine > RC Internal medicine
PhD Course: Cognitive, Computational and Social Neurosciences
Identification Number: https://doi.org/10.13118/imtlucca/e-theses/410
NBN Number: urn:nbn:it:imtlucca-30031
Date Deposited: 02 May 2024 13:50
URI: http://e-theses.imtlucca.it/id/eprint/410

Actions (login required, only for staff repository)

View Item View Item