Date of Award

7-16-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Engineering Technology

First Advisor

Mariofanna Milanova

Abstract

Eye gaze writing, a novel interaction modality, has the potential to revolutionize communication for individuals with limited mobility. In our research, we investigated the deep learning algorithms efficiency for real-time eye gaze writing. We have compared many algorithms' performance in many computer vision areas, such as object detection in which we used first YOLOv8, the second algorithm SSD, and the third algorithm is Faster R-CNN, the second computer vision area is the image segmentation in which we used DeepLab and U-Net, and the last computer vision area is self-supervised learning we have used SimCLR algorithm. By evaluating these models on our MPIIgaze data-set combined with the other five data-sets, we aim to identify the most suitable combination for accurate and efficient eye-gaze writing. Our findings demonstrate that Haar achieved the highest accuracy (0.85) with a compact size (97.358 KB), while YOLOv8 demonstrated competitive performance (0.83 accuracy) with exceptional speed and the smallest size (6.083 KB), making it ideal for real-time systems. The model that gives us the best accuracy with more speed will be selected and then integrated into a real-time application, enabling users to write text using only their eye movements. This research contributes to advancing assistive technology and opens up new possibilities for individuals with disabilities. By comparing YOLOv8, SSD, Faster R-CNN, DeepLab, U-Net, and SimCLR, we aim to identify the optimal combination for accurate and efficient text input. Furthermore, we explore the integration of text suggestion and correction features to enhance the user experience and improve writing speed. Our findings contribute to the advancement of assistive technology and open up new possibilities for individuals with disabilities.

Download

Share

COinS