Characterization of Visual Interaction within Eye Movement Data
by
Mohsen Farid
School of Computer Science, The Queen's University of Belfast
Coauthors: Uwe Kruger (School of Electronics and Electrical Engineering, The Queen’s University of Belfast) and Fionn Murtagh (Department of Computer Science, Royal Holloway, University of London)

To create a visual modality of interaction, one needs to understand the spatial distribution of eye gaze data as the eye fixates on objects of interest within a scene, specifically the dispersion of the (x,y) points of gaze from the point where the subject perceives himself to be looking at. Additionally there is the issue of performance in deciding whether the computed point of gaze is the proper point of gaze, which also requires proper pre-processing (filtering) of the data. In this regard the question arises as to how much data to collect in order to be enough to make such decisions.

To address the above needs, we analyze data collected during eye gaze sessions (1) to investigate of the interrelations between the movement along the x- and y-axes, and (2) to produce a dynamic model of the eye movement during fixation. We consider two fixation patterns: one dot of diameter 2 mm in the geometrical center of the computer screen; and a mesh of nine points spanning the whole screen. The latter is normally used as part of the equipment calibration. Data are collected at a sampling rate of 60 Hz, which is enough to provide a good account of the system dynamics.

The analysis presented here is conducted in a dynamic fashion and involves multivariate statistical tools, such as static and dynamic principal component analysis (PCA) and subspace identification. Such techniques are data-driven and projection based to enable the extraction of significant variation and to clip off insignificant and redundant information (Russel et al., 2000; Qin, 2003). The techniques used do not suffer from the same inherent problems as conventional system identification tools, as often reported in the research literature (MacGregor et al., 1991; Wise and Gallagher, 1996).

To enhance the data analysis, a filtering technique is introduced here with the aim of removing erroneous data points. Analyzing eye gaze trace data reveals that a total of four state variables are found, which account for accurate dynamic description of the equation of motion. This is consistent with the physiological description of eye movement for objects moving in a plane perpendicular to the visual axis.

References:

MacGregor, J.F.; Marlin, T.E.; Kresta, J.V.; Skagerberg, B. Multivariate statistical methods in process analysis and control. AIChE Symposium Proceedings of the 4th International Conference on Chemical Process Control, New York: AIChE Publ. No. P-67, 1991, 79.

Wise, B.M., Gallagher, N.B. The process chemometrics approach to process monitoring and fault detection. Journal of Process Control. 1996, 6(6), 329.

Qin, S.J., Statistical process monitoring: basics and beyond, Journal of Chemometrics, 17, 480-502, (2003

Russel, E.L., Chiang, L.H., and Braatz, R.D., Data-Driven Techniques for Fault Detection and Diagnosis in Chemical Processes, 2000, (Springer-Verlag, London)

Date received: March 29, 2005