Long-Term Eye Movement Analysis Using Wearable Electrooculography

Type: Semester thesis
Students: Michael Hardegger
Advisor: Andreas Bulling, Marc Bächlin
Project: Eye Movements and Visual Perception

In this thesis, an approach for automatic, context-based segmentation of eye movement data is described. Such a segmentation can for example be used for preprocessing of long-term data recordings for later classification, or as a support in annotation tasks. Experimental data was recorded using EOG and preprocessed based on existing code. The experimental design was restricted to typical office tasks. In the first part of this thesis, a set of eye movement features, which contain relevant context information for the experimental data, is proposed, as well as a method for extracting these features from EOG data. Numerous methods for segmentation based on these features were implemented and are described in this thesis. Evaluation showed, that particularly transitions between longer activity segments were correctly identified with recognition rates of more than 90%. The problem of detecting shorter distractions is discussed in a separate section. At the end of this thesis, optimal feature sets and methods for different applications are given.