Survey on Emotion Recognition Databases

Abstract

The social robots market has grown in recent years owing to advancements in robotics. Accordingly, there is a growing interest in emotion recognition technology, one of the fundamental technologies of social robots. Early research in emotion recognition predicted emotions into a Facial Action Coding System (FACS) that recognizes facial expressions by analyzing action units (AU), the fundamental action components of face muscles. However, with the advent of deep learning, models can now recognize facial emotions straight from images. For industry application of this deep learning based emotion recognition approach, supervised learning are usually applied, and this model training approach is highly dependent on training datasets. This survey gives an overview of datasets that can be used when the emotion recognition model is implemented. This paper focused on identifying the availability of datasets relevant to the elderly and children, the primary goals of social robots, as well as reviewing the data collection and annotation processes. This survey sheds insight on the current status of emotion datasets and presents suggestions for future development.