Investigating Classification Methods using Fixation Patterns to Predict Visual Tasks
Publication Date
1-1-2022
Document Type
Conference Proceeding
Publication Title
IFAC-PapersOnLine
Volume
55
Issue
29
DOI
10.1016/j.ifacol.2022.10.225
First Page
19
Last Page
24
Abstract
Studies have shown the possibility to classify user tasks from eye-movement data. We present a new way to determine the optimal model for different visual cognitive tasks using data that includes two types of visual search tasks, a visual exploration task, a blank screen task, and a task where a user needs to fixate at the center of any scene. We used CNN and SVM models on RGB images generated from fixation scan paths from these tasks. We also used AdaBoost on filtered eye movement data as a baseline. Our study shows that deep learning gives the best accuracy for classifying between visual search tasks but misclassified between visual search and visual exploration tasks. Machine learning-based methods performed with high accuracy classifying tasks that involve minimal visual search. Our study gives insight on the best model to choose by type of visual task using eye movement data.
Keywords
attention, classifications, CNN, Eye-movement, visual search
Department
Computer Science
Recommended Citation
Siddartha Thentu and Nada Attar. "Investigating Classification Methods using Fixation Patterns to Predict Visual Tasks" IFAC-PapersOnLine (2022): 19-24. https://doi.org/10.1016/j.ifacol.2022.10.225
