5.1 How is Eye-Gaze Interface Control Different?

One might ask why it is interesting or at all relevant to investigate the possibilities for eye-gaze control; we can do pretty well by using keyboards, trackballs, joysticks and mice, and our hands certainly seem more natural manipulating organs than do our eyes. Furthermore, interacting with the computer by eye-gaze could possibly be slower and put heavier cognitive demands on the user.

Yet, there is empirical evidence that using an eye-gaze based interface actually can be faster than traditional selection devices (Ware & Mikaelian 1987). This can readily be understood if one considers a slightly augmented version of the keystroke-level model put forward by Card et al. (1980). The keystroke-level model predicts the time to execute a unit task, T_execute, on the basis of the time for four physical-motor operators: H (homing-moving the hand to a different device), P (pointing-moving the pointer to the correct location), K (keystroking-performing a keystroke or mouse click), D (drawing a set of straight lines), a mental operator, M for mentally preparing operators, and a system response operator, R. The time for the response operator is determined by the computer system, whereas the remaining operators depend on the user and the layout of the screen and keyboard. Card et al. find that

The P-operator describes the operation of moving, say, the mouse to the correct location. It seems quite reasonable to subdivide this operator into an operator for moving your eye to the correct location, P_eye and an operator for moving the mouse to this location, P_hand. In some cases these two operations could overlap temporally-for example if one knows that one must click on an icon in the lower right corner of the screen and therefore initiates a downward motion of the mouse before the eyes fixate on the actual icon. In spite of this, it seems a reasonable approximation simply to add them together: P = P_eye + P_hand. Thus the time difference for mouse-operated and eye-gaze operated interfaces is described by:

Note that the homing operator, H, is only added if the user must move her hand from another input-device, an operator definitely not required in the case of eye-gaze control. In the case of latency-operated eye-gaze systems, where a selection is first executed when the user has fixated it for a set latency time, T_Reye-gaze cannot be less than this time. What the empirical evidence suggests, then, is that even with this time barrier we have that T_Phand(+T_H)+T_K+T_Rmouseclick> T_Reye-gaze, roughly speaking, that it takes longer to move the mouse and click on it than to wait for the set dwell-time. This is captured nicely in the phrase "the eye is there before the hand," indicating that normally we first fixate, and then move the hand to the fixated position. This makes eye-gaze based selection a more "direct" way of selecting displayed objects.

On the basis of the discussion about visual selective attention (section 3.2) one might think that selection by eye-gaze is more or less instantaneous, but oddly enough this is not so. The findings of Ware & Mikaelian (1987) suggest that Fitts' Law-which states that the time to perform a selection is T = c₁+ c₂ log₂(d/s + 0.5), where c₁ and c₂ are constants, d is the distance from the current point to the target and s is the size of the target-also holds for eye-gaze based object selection.

Thus we can conclude that eye-gaze control does have some of the traditional qualities of interaction (Fitts' law), but that it also is qualitatively different from traditional pointing, because some the operators usually performed are skipped, resulting in a faster selection time. Using eye-gaze control is a more "direct" way of interfacing with the computer.