Human Eye Gaze Estimation Algorithm Based On Head-mounted Eye Tracker

Head-mounted eye tracker is a new type of human-computer interaction devices which has been widely used in medical rehabilitation,augmented reality,and other fields.The basic principle is to capture the eye image by using gaze estimation technology and to establish the mapping relationship between the eye-movement features extracted from the image and the three-dimensional position of the gaze point to determine the actual gaze point or target object,so as to provide guidance for the subsequent virtual imaging or object grasping tasks.The existing eye gaze estimation algorithms based on head-mounted eye tracker are mostly proposed for the prediction of 2D screen gaze point.However,they are not suitable for guiding the robot arm to grasp the object in the actual physical environment.Furthermore,these algorithms depend on the complex structure of the eye tracker.The algorithm robustness and positioning precision are poor as well.Thus,this paper proposes a gaze estimation algorithm suitable for guiding the robot arm to grasp objects.This algorithm can accurately estimate the gaze direction of human eyes and the threedimensional position and orientation of target objects only depending on the eye tracker with simple structure.The main work content and research results of this paper contain:1)Realization of eye gaze direction estimation based on eye image.This paper analyzes the physiological structure of the eyeball and designs a simple eyeball model suitable for the estimation of gaze direction;Based on the model,a four-dimensional vector which can directly represent the gaze direction of human eyes is proposed,and then a gaze direction estimation network which can directly extract the four-dimensional vector from the eye image is designed;Finally,experiments are designed to verify that this model can estimate the gaze direction of human eyes accurately.2)Establishment of a mapping model to calculate the position of gaze points in the foreground image from gaze direction.This paper explores the mapping relationship between the four-dimensional vector representing the gaze direction of human eyes and the two-dimensional position of gaze points in the foreground image,and analyzes the advantages and disadvantages of the three common gaze estimation mapping models.Then the linear weighting method is used to fuse the quadratic polynomial regression model with the gaussian process regression model to design a new hybrid mapping model.Finally,three comparative tests were designed to verify the effectiveness and robustness of the model.3)Determination of the 3D position and orientation of the target object relative to the foreground camera.This paper first explores the mapping relationship between the location of the gaze point and target object;A set of target object pose reconstruction algorithm is designed to complete the determination of the 3D pose of the target object from the eye image.Finally,through two gaze estimation experiments for the object position in the real scene,we verified that the precision and robustness of the whole gaze estimation algorithm based on head-mounted eye tracker proposed in this paper.