Anonymous Detection Using Neural Networks for Smart Surveillance

This project presents a real-time surveillance system that intelligently detects weapons and anomalous human behaviour using deep learning models. The system integrates YOLOv8 for object detection to identify potential threats like knives, and ResNet50 for feature extraction from video frames. These features are then analyzed by an LSTM Autoencoder trained on normal behaviour to flag any unusual activity such as loitering, panic, or sudden erratic movements. A Flask-based API ensures real-time alerts and system communication, enabling quick response without compromising individual privacy. This approach enhances public safety by combining smart surveillance with AI-driven threat detection in dynamic environments. Keywords: Real-time Surveillance, Anomaly Detection, Weapon Detection, YOLOv8, ResNet50, LSTM Autoencoder, Deep Learning, Flask API.