Eye Model And Improved Itracker-Based Gaze Estimation

Gaze is an important non-verbal cue for speculating human's attention,which has deep and extensive applications in science,medical,business,education,criminal,human-computer interaction and other fields.With the development of microelectronics technology and digital image processing,more and more eye tracking and gaze estimation methods have been discovered.Although extensive gaze estimation methods have been explored,redundant calibration process,complex system settings,limitations of lighting conditions and the non-universal calibration for different subjects as well as low tolerance to headmovement remain challenges for the exsiting gaze estimation systems.In this paper,a novel iris center localization method is proposed to deal with the iris contour change caused by the rotation of eyeball using the geometric relationship among iris center,eyeball center and iris contour.Moreover,a novel 3D gaze estimation method is proposed by improving the Itracker and employing a many-to-one bidirectional LSTM(bi-LSTM)to fit the temporal information.A novel iris center localization method is proposed to deal with the iris contour change caused by the rotation of eyeball using the geometric relationship among iris center,eyeball center and iris contour.First,a face alignment method is employed to detect the face feature points that are used to initialize the iris center and eyeball locations.The three-dimensional rotation information of the eyeball is then described by the positional relationship between the centers of the eyeball and iris at the image level.The new obtained iris contour after the rotation is used to update the location of the iris center iteratively.Finally,experiments conducted on the BioID dataset and GI4 E dataset as well as eight subjects demonstrate the superior performance of the proposed method.We also propose an improved Itracker to predict the subject's gaze for a single image frame,as well as employ a many-to-one bidirectional Long Short-Term Memory(bi-LSTM)to fit the temporal information between frames to estimate gaze for video sequence.For single image frame gaze estimation,we improve the conventional Itracker by removing the face-grid and reducing one network branch via concatenating the two-eye region images.Experimental results show that our improved Itracker obtains 11.6% significant improvement over the state-of-the-art methods on MPIIGaze dataset and has robust estimation accuracy for different image resolutions under the premise of greatly reducing network complexity.For video sequence gaze estimation,by employing the bi-LSTM to fit the temporal information between frames,experimental results on EyeDiap dataset further demonstrate 3% accuracy improvement.