A Real-Time Autonomous Machine Learning System for Face Recognition Using Pre-Trained Convolutional Neural Networks

Abstract

Purpose: This paper aims to present a novel real-time, autonomous machine learning system for face recognition. This system employs pre-trained convolutional neural networks for encoding facial images and applies a Naive Multinomial Bayes model for autonomous learning and real-time classification. Design/Methodology/Approach: The system leverages a pre-trained ResNet50 model to encode facial images from a camera, while cognitive tracking agents collaborate with machine learning models to monitor the faces of multiple people. A novelty detection algorithm based on a Support Vector Machine (SVM) classifier checks whether a detected face is new or already recognized. The system autonomously starts the learning process if an unrecognized face is identified. Real-time classification of individuals relies on a Naive Multinomial Bayes model, with special agents tracking each face. Findings: Experiments demonstrated that the system can accurately learn new faces appearing within the camera frame in favorable conditions. The key determinant of successful recognition and learning is the novelty detection algorithm, which, if it fails, may assign multiple identities or group new individuals into existing clusters. Practical Implications: This system offers a practical solution for real-time, autonomous face recognition, with potential applications in security, access control, and personalized services. Its ability to quickly learn new faces while maintaining classification accuracy ensures adaptability in dynamic environments. Originality/Value: The research introduces an innovative approach by combining pretrained neural networks with autonomous learning and a novelty detection algorithm to classify faces in real-time. This hybrid method ensures rapid and accurate face recognition while minimizing the need for extensive training data or prolonged training times.