Vision Interaction Method Based on Visual Attention Mechanism

This paper puts forward a calibration-free vision interaction system using Visual Attention Mechanism (VAM), which requires the user to watch a video clip in advance to alleviate the problem of cumbersome and unnatural calibration, which exists in traditional gaze estimators. Our method establishes the mapping between eye features and gaze points using Support Vector Machine (SVM), in which eye features are constructed using Pupil Center Cornea Reflection (PCCR) vectors and gaze points are found using saliency maps. The experimental results show an increase in gaze estimator accuracy and and the vision interaction applications show the applicability of the vision interaction system.