Adaptive Work Mechanism Research For Eye-gaze Tracking Based Human-computer Interaction

With its immediacy, naturality and bidirectivity, eye-gaze based interactivemodality is of extensive application prospect in the field of Human-ComputerInteraction (HCI), assistance to the disabled and elderly, amusement games, driverassistance and virtual reality etc. Thus, eye-gaze tracking technique is of potentialdevelopment as a novel interaction method in HCI. And eye-gaze tracking technologybased on image processing is non-contact and non-invasive, which has attracted greatconcerns and achieved rapid development recently. However, imperfect yet as is theeye-gaze tracking technique, it is mainly applied to help the disabled and elderly withmovement disorders in the auxiliary functions and intelligent control systems.Nowadays, there are common drawbacks in most eye-gaze tracking system, such ascomplicate calibration procedure, restrictions on head movement, low robustness ineye-gaze tracking accuracy, low frame rate. Therefore, in order to settle theseproblems down, an adaptive eye-gaze tracking based human-computer interactionsolution has been provided in this paper, with following contents:(1) An eye feature location procedure is presented combining the methods offeature detecting with that of feature tracking. Firstly, a kind way of feature detectionand location is provided based infrared eye-image, which is time-consuming.Meanwhile, in order to make feature location process in real-time, optical flow basedfeature tracking method is proposed. In result, reliable and real-time is eye featuredetection and location workflow by integrating feature detecting process and featuretracking process.(2) To avoid the influence of head movement on eye-gaze estimation, a kind ofmulti-initialized states referred work parameter calibration is provided. In an eye-gazetracking system based on linear polynomial regression, eye-gaze directions are alwayscorresponded with the particular head position, which will be estimated wrong withhead movement. By taking the calibration parameters in multiple initial position asreference, work parameters will be corrected after automatic re-calibration procedure without interrupting the interaction process and repeating the initial complicatecalibration procedure.(3) A state-space model for eye-gaze tracking based on optical flow is established.In the multi-initialized states referred work parameter calibration mentioned above,the calibrating process is actually a course of solving polynomial regression equations.Thus, whether the calibration is succeeded or not depends on the presence of theequation solutions, which will bring issues on reliability and stability. Instead, thispaper analyzes the principle that eye-gaze estimation varies with the internal andexternal system states in light of the state-space approach method, and constructs astate-space model for eye-gaze tracking based on optical flow. In result, lesscalibration points are required and the calibration procedure is simple and fast.Therefore, a real-time eye-gaze tracking method is now offered, allowing headmovements.(4) The adaptive mechanism of eye-gaze based human-computer interaction isstudied. Firstly, human consciousness states detection and identification method isdiscussed, as well as the design and responding mechanism of the human-computerinterface, expanding horizons on improving the eye-gaze tracking basedhuman-computer interaction system into an adaptive one.