Emotion recognition in video using deep learning

Abstract

Affective computing aims to implement methods and technologies to recognize and synthesize human emotions. Understanding human facial expressions is essential to the success of this new branch of AI. Emotions can be conveyed through various channels, the most prominent are facial expressions, speech, texts and various other physiological signals. This topic has occupied researchers for a long time due to the difficulty of understanding and categorizing these expressions. In this work, we explore the different techniques carried out to recognize facial emotions in videos. We experiment on the AFEW dataset with two models based on deep learning. The first uses TCNs and the second uses CNNs. The experience with the first model was very hard since it belongs to recent sequential models and was not completed due to difficulty of implementation and limited resources. The second model achieved good accuracy of up to 91%.