Emotion Recognition using Facial Expressions in Children using the NAO Robot

Abstract

The detection of human emotions from facial expressions is crucial for social interaction. Therefore, several systems of behavioral computing in robotics try to recognize human emotion from images and video, but most of them are trained to classify emotions in adults only. Using the standard of 6 basic emotions: sadness, happiness, surprise, anger, disgust, and fear, we try to classify the facial expressions using the NAO robot in children. In this study, we make the comparison between the AFFDEX SDK, and a Convolution Neural Network (CNN) with Viola-Jones trained with the AffectNet dataset, and tuned with the NIMH-ChEF dataset using transfer learning to classify facial expressions in children. Then, we test our system comparing the CNN and the AFFDEX SDK for classification in the Child Affective Facial Expression (CAFE) dataset. Finally, we compare both systems using the NAO robot in a subset of the AM-FED and EmoReact datasets.