Virtual Booth

This study addresses the need for easily accessible, real-time, and non-invasive technologies to promote early identification and ongoing evaluation of psychological well-being by presenting the creation of an automated face expression recognition system for mental health monitoring. To improve model robustness and generalization, the system uses a large amount of data preprocessing and augmentation on the FER2013 dataset, which consists of 48×48-pixel grayscale photos from seven basic emotional categories. An EfficientNet-B0 architecture for effective feature extraction and Transformer-based attention mechanisms to capture intricate spatial connections were used to create a hybrid deep learning model. Techniques like learning rate scheduling, class balancing, focus loss, and early stopping were used to train and optimize the model. Classification metrics and confusion matrices were used to assess the model's performance. A validated self-assessment questionnaire (influenced by PHQ-9 and GAD-7) is also included in the system, which is integrated into a real-time interface using OpenCV to provide a thorough assessment of users' emotional and mental states. The suggested method shows promise for practical mental health applications, particularly in situations that call for continuous emotion monitoring and passive evaluation of psychological health.