Advancements, Methods, and Obstacles in Emotion Recognition with Multimodal Approaches

Emotion recognition has gained an avid curiosity in the research area due to its broad range of applications in human-computer interaction particularly in real time emotion recognition, sentiment analysis, and affective computing. Deep learning techniques are advancing exponentially, automating feature engineering and transcending traditional machine learning techniques. Appeal of the above models can be highly beneficial but also raise certain technical and ethical concerns that are not often discussed enough in the existing studies. While highlighting the recent advancements, the methods involved and obstacles in emotion recognition, this review also proposes future directions in developing a standard unified framework for multimodal approaches. Categorizing and analysing the recent studies depends upon the model types employed, data modalities used and the type of fusion techniques used, this review was able to gain insights into the effect of recent multimodal approaches addressing technical and ethical concerns. Utilization of fusion techniques in multimodal models is still not reliable but they better convey the connection between different modalities and have a huge potential for development in the future. This review provides insights into the state of art multimodal methods using modalities like facial expressions, body gestures, vocal cues, text, and physiological signals for researchers to understand the trends and future directions in this field.