EVAI: An Approach for Real-time Video-based Emotion Recognition

Facial expression recognition is the basis of emotion recognition. The expression recognition algorithm’s inevitable result jitter hampers the continuous observation and judgment of emotions. The purpose of this paper is to develop a facial expression recognition task with high precision and real-time characteristics based on the YOLOv4-tiny network architecture, by proposing a quantitative Emotion Variation Amplitude Indicator (EVAI) to quantify the rating of expressions, stabilizing the outputs of the recognition results as much as possible using a frame-by-frame optimization method, which improves emotion recognition accuracy.