Publication

Abstract : This research revolutionizes real-time video analysis by seamlessly integrating emotion, age, gender, and object detection in each frame of a live video feed. The framework leverages a unified deep model that combines pre-trained feature extraction models with CNN architecture, each meticulously trained on specialized datasets tailored to its recognition task. The proposed model achieves commendable accuracy of 85.28% in emotion recognition using FER-2013 dataset, 84% in age estimation, and 88.67% in gender classification using UTK-Face dataset, and 90.81% in object recognition leveraging the Fruits and Vegetables dataset. Moreover, the emotion, gender and object models achieved an average precision of 0.865, recall of 0.866 and F-score of 0.864, while age model achieved a mean squared error of 0.1136, quantitatively showcasing the model performance.