Real-time emotion analysis: Intelligent system design based on deep learning

Understanding facial expressions, which serve as the key to understanding emotions such as emotion recognition and behavioral judgment, is very valuable in people's daily lives. Although it is easy and insignificant for people's eyes, it is difficult for machines to perceive it accurately and requires many image processing techniques to be applied. Facial emotion recognition systems, which have an important place today, are being applied in various areas and continue to attract attention. Since the prediction of facial emotion with age and gender in live camera captures will serve many potential purposes, in this study, the stress rate is calculated based on the emotional perspectives of the faces in the live camera image. In the study, FER2013, FER+, and UTKFace were used as data sets, and customized CNN model, VGG-16, ResNet-50, and ResNet-152 architectures were preferred among CNN models. In the obtained results, the ResNet-50 model provided the best accuracy performance. All predicted instant situations in real life are able to help in various application areas such as education, health, job security, crime detection, trade, etc. Although many systems that are currently applied exist in daily life, we have completed this study in a way that has not been in the literature before, aiming to take these systems one step further by gathering these systems under one roof at the same time.