Anomaly Detection in Crowds

A deep learning pipeline for detecting abnormal crowd behavior in public spaces. The system trains a convolutional neural network on historical crowd footage to learn what normal activity looks like, then scores incoming frames against that learned baseline in real time. Anomalies are surfaced with configurable sensitivity thresholds and visualized as frame-level heatmaps that highlight exactly where unusual activity is occurring, making it actionable for safety operators rather than just a binary alert.