CERDL: Contextual Emotion Recognition Analysis Using Deep Learning.

This paper delves into the critical importance of understanding emotions from a person’s perspective, and the potential for machines to improve human interaction by possessing this ability. While existing research on emotion recognition in computer vision has mainly focused on analyzing facial expressions and categorizing them into six basic emotions, it is important to recognize that contextual factors also play a crucial role in emotion perception. Emotions are not just limited to facial expressions but also include body language, the pitch of voice, and other nonverbal cues. We then trained a convolutional neural network model on this vast dataset and demonstrated the importance of incorporating context to recognize rich information about emotional states in images. Our model surpasses previous benchmarks and confirms the value of contextual information in emotion recognition. We have used the Emotions in Context (EMOTIC) [1] and Body Language Dataset (BoLD) [2] datasets for recognizing emotions by taking their contextual information into account. By incorporating contextual factors, machines can enhance human interaction by accurately recognizing emotional states in various situations. Based on the experiments, we recognized that the emotions of engagement (93.62%), confidence (92.41%), and excitement (95.93%) were predicted accurately. In contrast, the emotions of yearning, disapproval, and pain had low classification accuracy, with less than 40% accuracy. Lastly, this paper highlights the importance of understanding emotions beyond just facial expressions and provides a benchmark for emotion recognition in a contextual setting.