Gaze Estimation Based on Single Objective Natural Conditions

Abstract:
Appearance based gaze estimation has great application value in the field of human-computer interaction. Nowadays, most deep learning methods require a lot of calculation and time. In this experiment, I designed a new mapping algorithm from three-dimensional coordinates to two-dimensional coordinates, and made the fixation point control the mouse movement to achieve the function of human-computer interaction. When the value 1 represents the maximum accuracy, the fixation accuracy of this experiment can be maintained above 0.8 when the head posture is not moving, High accuracy can also be maintained in the case of head posture movement. Compared with the depth learning method, our accuracy is equivalent, and compared with the traditional method, our accuracy is better. Moreover, the method designed in this paper has high fixation mapping efficiency and lower equipment requirements. The method proposed in this paper has great prospects in human eye control computer interaction.