Real-time emotion recognition based on facial expressions using Artificial Intelligence techniques: A review and future directions

Abstract

In recent years, the real-time facial expression recognition system based on artificial intelligence technology has garnered significant attention from academia and industry. This paper presents a systematic literature review and bibliometric analysis to examine the latest publications in this field, summarizing the development and research significance of facial expression recognition technology and emphasizing its vital role in human-computer interaction and affective computing. The study used PRISMA to review 386 articles published from January 2019 to December 2023 in Web of Science, Scopus, IEEE Xplore, and ACM Digital Library. It encompasses covering various research methodologies, datasets, and application areas, as well as artificial intelligence technology, algorithms, and models. This review highlights advancements in Facial Expression Recognition, particularly the predominant use of databases such as FER2013 and CK+ while identifying Convolutional Neural Networks as the primary technique for real-time emotion classification. A quantitative analysis of research trends over the past five years indicates a shift toward keywords like transfer learning and applications in domains such as healthcare and the Internet of Things. Contemporary deep learning models, including CNNs, ResNet, and VGG, demonstrate impressive accuracy in classifying seven basic emotions, facilitating real-time applications across multiple fields. However, challenges such as overfitting, sensitivity to environmental factors, and the necessity for high-performance computing resources impede the broader deployment of these systems. These findings underscore the urgent need for further research to address these limitations and enhance the ethical application of FER technologies. Finally, based on the review and analysis results, this paper outlines future research directions for this technology, including multimodal information fusion, computational modelling, personalized emotion recognition, and interdisciplinary cooperation, thereby providing valuable references and inspiration for future works.