Real-Time Age and Gender Detection System Using Deep Learning

Abstract

Age and gender detection in computer vision is key for real-time human analysis via video streams. This project employs Python, OpenCV, and Caffe deep learning models to create a system that captures live video through a webcam, detects faces using a Haar Cascade classifier, and predicts age group and gender using pre-trained Caffe classification networks. The system was evaluated on the Adience benchmark test split, achieving a gender classification accuracy of 86.4% and an age-group classification accuracy of 79.1%, with an average processing speed of 18-22 frames per second (FPS) on a CPU-only laptop and 46-52 FPS on a GPU-enabled system. Image pre-processing techniques enhance accuracy, and the system operates efficiently on standard hardware under favourable conditions. Experiment results indicate that factors such as lighting and facial pose affect prediction accuracy. Despite challenges, the system produces user-friendly age-group labels.