Emotion Recognition at Real-Time Applications using Meta-Learning

Emotion recognition is one of the important applications of computer vision and artificial intelligence for human computer interaction. ‘Surprise, Anger, Neutral, Fear, Sad, Happy, and Disgust’ are the seven basic emotions of humans. In this work, CMU MultiPIE database has been used to train and test the emotion recognition model. This model is then used to test the emotions of people in real-time in situations like interview, online classroom, office meetings, other human interactions etc. The emotions considered in this work are happy, surprise, angry, disgust and neutral. In interviews, a face-scanning algorithm helps to decide whether one deserves the job or not. Moving beyond recruitment, the same model can be used in schools to assess whether students are paying attention or to identify whether employees are paying attention in office meetings. It is required to develop a model that can detect emotions in videos with less number of images in the training set that works well for unknown samples. Meta-Learning provides a solution for this problem. The train database consists of front pose face images from CMU MultiPIE database. The test database contains various head pose images from CMU MultiPIE database. The proposed model provides an accuracy of 95% and 85% for training and testing respectively.