K. Horii1, K. Kotani1, Y. Kitamura2, G. d'Ydewalle3
1Department of Industrial Engineering, Kansai University,
Japan (e-mail:khorii@iecs.kansai-u.ac.jp);
2Department of Informatics, Kansai University, Japan;
3Department of Experimental Psychology, University of Leuven,
Belgium
The authors have been working on eye movement in reading text, and in this study it is vital to achieve accuracy in the determination of fixation as the measurement of fixation duration seems to have strong relationship with evaluating cognitive processes involved in reading. A great variety of the techniques for determining fixation has been proposed and yet none of them are free from the limitation imposed by the sampling rate and the accuracy of the eye movement recording device that is employed for the eye movement studies. Traditionally, fixations have been determined by the co-ordinates of loci and the time threshold when a cluster of the co-ordinates stayed within a certain radius. Some researchers may use topological techniques to cluster out groups of co-ordinates into fixations. The authors have long awaited for a day when we can employ the velocity and the acceleration of saccadic eye movements for the detection of fixation since the onset of a fixation is marked with the end of a saccade and the end of a fixation is marked with the onset of a next saccade.
For making this come true, NAC's EMR 7 that runs at 30 Hz is far from being adequate in terms of sampling resolution. EMR 6 (600 Hz) seemed to be what we needed but it turned out to fall short of our expectation because the data obtained was more distorted than being negligible, which could not be compensated by data processing. At long last, we obtained Eye Link System (250 Hz) and we developed the programs for data acquisition and analysis. With this utility software the system proved to be accurate enough to detect not only the onset of a saccade when the eyeball already has a certain amount of inertia before triggering a saccadic movement but also it can cope with those saccades when the eye overshoots. The system can differentiate the real saccades from 'phantom' saccades that sporadically appear probably on account of sampling noise and/or tremors during smooth pursuit.
Our research result using Eye Link System is that the onset of a saccade can best be detected to be the point m when the acceleration is over 1,300 deg/s2 at m + 1 point, 3,000 deg/s2 at m + 2 point, and 5,000 deg/s2 at m + 3 point, and that the saccades terminates at the point n when the acceleration falls below 2,000 deg/s2 and at n + 1 point the velocity is below 100 deg/s as well as the acceleration is below that of n point.