This thesis investigates several problems related to eye tracking, including computing a mean gaze path from multiple subjects' eye movements and detecting visual attention without calibration.
To compute a mean gaze path, the thesis develops two methods: 1) A method based on feature networks that produces a graph representation, and 2) A more shape-oriented method that further reduces the data and produces trajectories.
To detect visual attention without calibration, the thesis develops two methods: 1) Using hidden Markov models to passively identify saccade patterns from stimuli, and 2) A smooth pursuit comparison and performance evaluation method.
Both methods for detecting visual attention aim to determine what eye movements a subject is capable of during visual attention in order to