Research On Gaze Estimation Algorithm Based On Eye Movement Features And Head Pose

With the continuous development of science and technology,people’s research in the field of human-computer interaction is also getting deeper and deeper.The humancomputer interaction based on the non-physical contact of human eyes provides convenience for the disabled.At present,this research has been widely used in medical,psychology,virtual reality and other fields.Most of the current gaze point estimation methods will restrict people’s head posture,require a very strict experimental environment,and cannot meet people’s daily use,and most use highprecision hardware devices to ensure the accuracy of gaze point estimation,resulting in overhead problems such as large size and difficulties in practical application.This paper studies and designs a method of gaze estimation under various unconstrained conditions,such as natural light,head posture,no specific experimental environment,and no need for high-precision equipment.Only one camera equipped with a computer is used for experimental research,and based on this,the information extraction of eye posture features and the estimation of head posture are carried out.The combination of the two is used for correction to obtain the position of the gaze point mapped on the screen of the experimental equipment.The main work of this paper is as follows:1.Eye movement feature information extraction:Face detection is performed on the input image,and part of the human eye area is segmented.Perform edge detection on the extracted eye feature image,use gradient strength detection and quadratic circle fitting to fine-tune the center point of the iris,and use the Harris corner detection algorithm to extract the inner corner of the eye.These two eye feature information The extraction greatly improves the accuracy and stability of the experiment,ensures the real-time performance of the experiment,improves the adaptability of the experiment to natural lighting,and enables the algorithm to be applied under natural lighting.2.Head pose estimation:The experiment extracts the features of the head pose,and uses the mapping relationship between the three-dimensional coordinate points and the two-dimensional feature points to obtain the parameters of the head pose.Use the Euler angles to calculate the pitch of the head rotation around the X axis,yaw around the Y axis,and roll around the Z axis to obtain the head pose and optimize the head pose solution.3.Gaze point coordinate positioning:Select a polynomial gaze point model,calibrate the model to obtain relevant parameters,and combine the coordinates of the iris center point and the inner corner of the eye to initially locate the gaze point position.Considering the influence of head posture on gaze point estimation,a head posture compensation algorithm is proposed,which combines the deflection angles of the head in three directions to compensate the error caused by head movement,and realizes the head movement under the condition of head movement.The flexibility and real-time performance of the user’s use allow more precise positioning of the gaze point.