Application of Deterministic and Stochastic Components of the Ocular Dynamic System.

摘要

With the advent of eye tracking technology, gaze control is perceived to be an alternative way of human-machine interaction. Owing to the nature of eye movement, visually localizing a target is faster than by hand control and less likely to cause fatigue. Because of the connection between eye movement and attention, it is plausible to infer what targets may be of interest to people based on their gaze pattern. However, current design paradigms focus on fitting eye movements to user interfaces initially designed for hand movement. For example, many studies attempted to utilize gaze point as a mouse cursor, which resulted in limited success. To date, gaze-based interaction has been prone to mistakes and it is time consuming compared to other conventional non-keyboard input methods. Therefore, commercial eye tracking systems are almost exclusively used by people with severe disabilities or for military purposes.;Although eye movement and hand movement share biomechanical similarities, there are fundamental differences between the two which pose difficulties encountered in gaze-based interaction. First, eye movement is involved in both active interactions and passive information perception. Looking at a menu item indicates the user "probably", but not necessarily, intends to click on it. Hand movement on the other hand, serves solely as an interaction proxy without ambiguity. Secondary, from the view of control mechanisms, eye movement is governed by open loop control whereas hand movement is closed loop control with visual feedback. All the models developed for goal-directed hand movement assume a feedback control mechanism, which does not apply to eye movement. These models play a key role in understanding the underlying control mechanism and the speed accuracy tradeoff of hand movement. Specifically, motion time is a function of target distance and target size (i.e., Fitts' Law). Due to the fact that closed loop control is not valid for eye movement, previous models for hand movement are no longer appropriate for explaining dynamic features of the ocular plant.;By addressing these two differences between eye movement and hand movement, this dissertation aimed to improve the usability of eye tracking interaction and extend the existing goal-directed model for hand movement to three-dimensional eye movement. A prototype interface called the "Hot-Zone" was implemented for gaze-based interaction and assessment. Previous studies focused on utilizing gaze control for one-stage control such as 'click' on an existing interface. The Hot-Zone acts like a context menu designed for gaze, which provides a solution to two-stage control, such as "Call out first, then select". An experiment was conducted to show the performance of this technique compared with a conventional pointing device.;Because eye movement is closely related to information processing, it is possible to infer the task context (e.g., browse a picture on the web or read an article) by analyzing eye movement data. A system with predictive intelligence to evaluate a user's attention based on passive eye movement could be beneficial to support interaction quality and improve productivity. Hidden Markov Models and Support Vector Machine were used to differentiate reading and visual searching gaze patterns based on eye movement.;To explore the control mechanism of eye movement, deterministic and stochastic components of the ocular dynamic system were explicitly constructed based on mechanical principles and neuronal and physiological evidence established so far. The kinematics were verified by simulation and compared with experimental data for validation. This simulation model extends the eye movement from one-dimensional to three-dimensional space. Supplied with a velocity profile as input, the model applies stochastic components during the movement and predicts the final state of gaze. In order to generate the velocity profile with reasonable credibility, an optimal control model is developed. This optimal control model reveals a plausible criterion, which governs the eye movement. The criterion explicitly shows the tradeoff between the competing considerations (e.g., time optimal and minimum effort) in motion planning. Being a deterministic model by its nature, the present model resembles the minimum variance model numerically from the view of optimization. The latter was considered to be one of the few models capable of accounting for stochastic components of eye movement, as it addresses signal-dependent noise.