Eye Tracking in Human-computer Interaction Recognition

Eye tracking interaction has a wide range of application prospects in the field of human-computer interaction, and it is generally invasive, the calibration process is complex and expensive, and the resolution of ordinary monocular camera sensors is low. In this paper, an eye movement behavior recognition method based on front-facing camera video source using directional gradient histogram (HOG) feature + SVM+LSTM neural network is proposed, and then a simple human-computer interaction application is realized. Firstly, the face is located and tracked, the binocular area is obtained according to the coordinates of the four key points at the corner of the eye after the face alignment operation, the SVM model is used to determine the closed and non-blinking states of the eyes, and then the position of the center of the eye between adjacent frames is analyzed to roughly judge the eye movement, and the suspected intentional eye potential differential video sequence between frames is input into the LSTM network for prediction, and the eye movement behavior recognition results are output, and then the computer command is triggered to complete the interaction. After testing 20 000 samples (about 10% of which were negative samples) in the self-made data sample set, the accuracy of dynamic blink recognition and 99.3% of eye movement behavior prediction was better than 95%.