Gaze-Estimation-Based-on-Deep-Learning-and-Mathematical-Method

Gaze direction can be defined by the pupil and the eyeball center where the latter is unobservable in 2D images. Hence, achieving highly accurate gaze estimates is an ill-posed problem.Therefore, I tried to extract several efficient landmarks around eyeball and iris for gaze estimation from single eye input. Instead of directly regressing two angles for the pitch and yaw of the eyeball, I regress to an intermediate pictorial representation which in turn simplifies the task of 3D gaze direction estimation.I tried to use a novel learning-based method for eye region landmark localization that enables conventional methods to be competitive to latest appearance-based methods. In the meantime, I refer to the extraction network of human body feature points, designing a totally new net to improve my results. My approach outperforms existing model fitting and appearance-based methods in the context of person-independent and personalized gaze estimation.