At present, there are a variety of eye signals that can be used for emotion recognition. General eye information such as fixation time, pupil diameter, saccade time and other data need to be collected by using a professional eye tracker. This paper proposed an emotion recognition method based on pupil position, which used gradient-based pupil positioning algorithm to directly obtain pupil position coordinate data from facial video. For the preprocessed data, it proposed the comprehensive waveform complexity as the feature value. By segmenting the coordinate waveform of the pupil position, it obtained the sum of the absolute values of the correlation coefficients between all segments, thereby obtaining the comprehensive complexity of the waveform degree. Experimental results show that the pupil position coordinates can be used for emotion recognition, and for the pupil position coordinate waveform, the effect of comprehensive waveform complexity feature value emotion recognition is better than sample entropy and Lemple-Ziv complexity. In the future, the pupil position coordinates can be used together with other physiological signals to improve the accuracy of emotion recognition.

[2021-06-05] Printed Article /doc/2020.06.0165/v2