Affective Computing in Computer Vision: A Study on Facial Expression Recognition

The use of artificial intelligence has become increasingly popular in recent years, allowing technology once thought of as futuristic to become possible and utilised at the consumer level. Many technological barriers to human-computer interaction have been overcome, and there is now a focus on the sociological acceptance of such technology. Inferring human emotional states is a time-consuming process and can be automated with computer vision. In this study, we explore how computer vision and face recognition systems can be leveraged to automatically infer human emotional states from the face. Rather than the classical single-emotion classification method, our aim is to explore whether it is possible to perform regression techniques to observe valence and arousal. Following the topology tuning of 33 different neural networks, the results show that valence and arousal can be predicted by a branched Convolutional Neural Network model with a mean squared error of 0.066 and 0.107, respectively. In addition, we discuss methods of improving the model, as well as uses of the technology, which include the autonomous monitoring of affect during situations of technological acceptance.