Dynamic Gender Recognition and Emotion Analysis in Live Video Feeds

Abstract

This study presents a video-based facial analysis framework for real-time age, gender, emotion, and ethnicity recognition using convolutional neural networks (CNNs). Our system employs a custom cascade classifier for face detection and utilizes four distinct pre-trained CNN models optimized for each respective classification task. The proposed model achieves an age prediction accuracy of 88.5%, gender prediction accuracy of 94.2%, emotion recognition accuracy of 85.1%, and ethnicity classification accuracy of 90.3%. Additionally, our approach maintains a processing speed of 15 frames per second on a standard desktop GPU, ensuring seamless real-time performance in live video feeds. These metrics underscore the reliability and practicality of our system for applications in human-computer interaction, surveillance, and demographic studies.