FEASIBILITY OF A REAL-TIME MOBILE EMOTION RECOGNITION SYSTEM FOR CHILDREN

This feasibility study focused on developing a robust Facial Emotion Recognition (FER) system specifically tailored for children aged 7 to 12, aiming to address the limited availability of specialized tools in this field. The system development followed the standard FER pipeline, including face detection, feature extraction, emotion classification, and evaluation. To determine the most effective classification method, three deep learning models, DenseNet-201, ResNet-101, and Inception-v3, were trained using MATLAB. The results showed that DenseNet-201 yielded the highest overall accuracy at 63.10% and was subsequently chosen for integration into the mobile system. The developed system was able to analyze facial expressions frame by frame, offering accurate insights into the emotional states of children. It also supports a user-initiated video processing workflow through an intuitive user interface (UI), minimizing user effort while strictly adhering to privacy protocols.