Real-Time Person Recognition and Localisation System for Emergency Management on Large Cruise Ships

Abstract

Safety in engine rooms on cruise ships is a challenge due to complex layouts, high numbers of people and extensive supply systems. Emergencies such as steam pipe bursts can severely restrict visibility, causing delays in evacuation in hazardous areas. This delays the detection of people in need of assistance and increases the risk of injury. We present an AI-based proof of concept for real-time person detection and localisation. This approach uses surveillance cameras to detect people and visualise their locations on a map with predefined ship areas. Temporal tracking ensures an up-to-date overview of the distribution of people. The system was tested in a simulated indoor environment. Raspberry Pi cameras, synchronised via the Network Time Protocol (NTP), ensure uniform time stamps. Overlapping camera areas were optimised to avoid multiple counts. The pre-trained real-time model YOLOv8 was used for person recognition. The locations were visualised on a model map using OpenCV and Matplotlib. Future work will focus on improved robustness in poor visibility, scalability for larger surveillance areas and optimised modelling. Beyond cruise ships, the system offers potential for other safety-critical areas with high passenger density and complex infrastructures. Precise real-time tracking facilitates and accelerates rescue operations and increases safety standards for passengers and staff.