Custom Convolutional Neural Network (CNN) Model for Emotion Recognition in Real-Time Video

Abstract

Facial Expressions Recognition (FER) is a vital task in human-computer interaction. This field can be used to augment many other fields such as learning, content suggestions and many more. This paper aims to propose a simple FER approach utilizing a custom Convolutional Neural Network (CNN) architecture for real-time emotion recognition, complemented by Hard Cascades for efficient face detection. The model is trained on a large and reputable dataset encompassing various facial expressions classified into 7 main emotions. The results of this study yields an accuracy of 60.78% achieved on the training data and 57.32% on the validation data. This shows that with even a very simplified approach with a custom CNN model, we are able to relatively accurately recognize facial expressions. This study aims to be a kind of groundwork for more advanced approaches to FER.