Real-time Criminal Face Recognition Using Machine Learning

This paper considers the relative performance of three popular machine learning models: Convolutional Neural Networks, Recurrent Neural Networks, and YOLOv8 (You Only Look Once version 8), in the process of real-time criminal face recognition. Such correct identification of someone in real-time live video feeds can be of much help to law enforcement in this high-tech security world. Comparing the strengths and weaknesses of every one of these models, CNNs are great for image feature extraction, RNNs good for the need to process sequences for temporal analysis, and YOLOv8 is good for the speed of object detection. The evaluation uses the performance metric-based accuracy, precision, recall, and F1-score. Distance measures, such as Manhattan and Euclidean distances, will be implemented for comparison in face recognition. The outcome of this research shows that YOLOv8 has been shown to process the fastest, while CNNs have proven better at recognition tasks and RNNs improved model recognition of temporal context. This work on criminal face recognition requires optimization in selecting apt models for enhancing the power of real-time surveillance in matters of safety issues in the public sphere.