Statistical analysis of eye gaze data.

摘要

In this dissertation, we present analysis of eye gaze data in response to simple one-arm movements. We report the presence of long memory property in the eye gaze. Using this property, we model the eye gaze data. The best model involves present arm coordinates as well as lagged eye gaze and arm coordinates. Further analysis for classification of eye gaze under two conditions "Watch" and "Imitate" is presented. This analysis uses time domain and spectral domain methods. In the time domain, properties of higher order crossing (HOC) sequence is investigated for long memory processes. Further, we present its application to eye gaze data. The application reveals differences in the two conditions for some of the subjects. In the spectral domain, logarithmic ratio of spectral densities are assumed to follow either the exponential (EXP) or fractional exponential (FEXP) model form. By applying assumptions on the distribution of periodogram ordinates of long memory processes and using FEXP model, classification of long memory processes are studied using simulations. The results are compared when EXP models are also used. Both techniques are then applied for classification of eye gaze data.