A Lightweight Hybrid System for Crowd Stampede Prediction Using UAV-Based SSIM Analysis and RFID Sensor Fusion

Crowd stampedes continue to be a serious concern during large gatherings, mainly because they occur suddenly and are difficult to foresee. In this project, I worked on developing a simple system that can give an early indication of when such a situation is beginning to form. The approach brings together three kinds of inputs: live aerial footage from a UAV, a few reference images taken from known stampede events, and basic movement information collected through RFID tags worn by people in the crowd. Each incoming video frame is checked against the reference images, and whenever the similarity appears unusually high, the RFID readings are examined to understand how the crowd is behaving at that moment. Using these readings, the system calculates a probability score and then converts it into an entropy value to estimate how stable or unstable the situation is. If the entropy becomes too low or too high, the system triggers a warning by activating a hooter so that the authorities can respond before the situation worsens. By combining very lightweight visual comparison with simple sensor data, the method aims to offer a practical and fast way to detect the early signs of a potential stampede.