Research On Human Eye Gazing Point Estimation

Gazing estimation is a research subject built upon machine vision and image processing. Interaction through eye movement is direct and natural. As an effective tool to measure human behavior of consciousness, eye gaze tracking technology has been widely concerned by researchers from a lot of fields such as cognitive neuroscience, psychology, industrial engineering, marketing and computer science. This dissertation addresses the issue of face detection, eye localization, feature extraction, system design, and system set up by summarizing the research fruits achieved both at home and abroad. The major content and contributions are as follows:1. A face detection method is proposed based on the combination of Adaptive Boost (AdaBoost) and Particle Filter (PF), which provides a solution for face detection in complex background. With the help of AdaBoost, it is no longer a necessary condition to have manually given initial position of target as required by PF, and meanwhile PF overcomes the problems of false detection and missing detection that exist with AdaBoost method. So the combination of AdaBoost and PF produces an effective complementary effect.2. A human eye localizaion method is proposed based on face image. Firstly, models of face shape and texture are built by active appearance model (AAM) globally. Then, taking advantage of different convergence characteristics of MVLR (Multivariate Linear Regression) and ICIA (Inverse Compositional Image Alignment), in the process of parameters fitting, the parameters of the global transform are estimated by MVLR and face shape parameters are estimated by ICIA, so that an accurate convergence result is obtained, whereby realizing a precise eye location finally.3. An eye gaze estimation method is proposed based on twice polynomial fitting. The first polynomial fitting is used to study the functional relation between the relative offset of iris center and the coordinate of reflected light spot when gazing point is fixed. In the second polynomial fitting, study is made on the mapping relationship between the relative offset of iris center and gazing point when the position of reflected light spot is determined. Therefore, eye gazing point can be effectively estimated through twice polynomial fitting.4. An eye gaze estimation system is designed and built based on infrared light source, and in the system calibration process, a method is suggested on how to measure and calibrate the relative position of auxiliary light source in the system, which provides a reliable hardware environment for an accurate eye gaze estimation.