Research On 3D Gaze Estimation Methods Based On Simplified Hardware System

Gaze tracking uses various existing detection methods such as machinery,electronics,and optics to obtain the current "gaze direction" of subjects.It has been widely used in human-computer interaction,medical diagnosis,national defense and military,traffic safety,human factor analysis,virtual reality and other fields.In recent years,the rapid development of new-generation information technologies such as big data and artificial intelligence has also put forward higher demands for gaze tracking technology.It has its limitations whether from the perspective of theoretical research or practical application.The current gaze tracking model mainly has the following three technical bottlenecks:(1)It is often necessary to obtain user-specific feature parameters through the user calibration process.(2)It restricts the natural head movement.(3)The research on simplifying user calibration and allowing head movement is mainly at the cost of the complexity of the hardware system.In addition,the application of gaze tracking technology in portable electronic devices such as mobile phones and tablet computers can be abstracted into gaze estimation under a simple hardware system configuration.However,the current simple hardware system configuration can only achieve 2D gaze estimation,and there are problems such as complicated user calibration and restricted head movement.In comparison,3D gaze estimation can simplify the user calibration process and allow head movement.Therefore,in-depth research on 3D gaze estimation methods based on simplified hardware systems,solves the problem of complex system configuration of 3D gaze tracking system,and improves the operability and usability of gaze tracking systems,is the basis for realizing simplified user calibration,full freedom and large-scale head movement.It is necessary and has practical significance to make the gaze tracking technology a real natural human-computer interaction or human factor analysis method and be applied to portable electronic devices such as mobile phones and tablet computers.The main contributions of this thesis include:(1)Aiming at the problem that the existing 3D gaze estimation method based on a single-camera-single-light-source system needs to preset some eye invariant parameters,an iris-radius-calibration based 3D gaze estimation method for the single-camera-single-light-source system is proposed.The 3D reconstruction of the spatial iris center and its normal vector is used,which realizes the iris radius calibration and 3D gaze estimation under the system configuration of single camera and single light source.The 3D line-of-sight(LOS)is then optimized based on the binocular model.This method solves the problem that the existing 3D gaze estimation method requires a single-camera-multi-light-source system or a multi-camera-multi-light-source system to estimate the 3D LOS,and breaks through the limitations of existing methods on hardware system configuration.(2)On the basis of-simplifying the hardware system configuration required for 3D gaze estimation and breaking through the limitations of existing methods on the hardware system configuration,aiming at the problem that the 3D reconstruction of the spatial iris normal vector is greatly affected by the iris imaging feature parameter error,a cornea-radius-calibration based 3D gaze estimation method for the single-camera-single-light-source system is proposed.After calibrating the iris radius,the cornea radius is then calibrated based on the light source imaging model in the user calibration process,which considers individual differences.In the process of 3D gaze estimation,the cornea center is calculated according to the spatial geometric model,then the cornea center and the iris center are used to reconstruct the optical axis(OA)of the eyeball,estimate and optimize the 3D LOS,which avoids the OA error introduced by the inaccurate iris normal vector.(3)For the typical single-camera-two-light-source system in which the two light sources are collinear with the camera optical center,the 3D gaze cannot be estimated using the existing method for the single-camera-two-light-source system when the invariant parameters of the human eye are unknown.To solve this problem,an iris-feature based 3D gaze estimation method for the typical single-camera-two-light-source system is proposed.Firstly,the cornea radius is calibrated using the iris features,and then the 3D LOS is estimated using the iris features or pupil features.This method makes it possible to realize 3D gaze estimation in this typical single-camera-two-light-source system.