Dynamic UAV-based search operations using probabilistic diffusion modeling of Man Overboard incident victims
2026-08-03 • Robotics
Robotics
AI summaryⓘ
The authors studied how to predict where a person who fell overboard from a cruise ship might be, to help find them faster. They used a computer model called the Extended Kalman Filter along with information about wind, water currents, and how people drift in water to estimate the search area. They then tested five different drone flying patterns to find the person, with one method called Improved Probability Informed Search working best. This method found people over 80% of the time, even if the search started 20 minutes after they went overboard. The authors also shared their code publicly for others to use.
Man OverboardExtended Kalman FilterLeeway ModelUnmanned Aerial Vehicle (UAV)Search and RescueProbability Informed SearchDrift PredictionSearch PatternsDetection ProbabilitySimulation
Authors
Dimosthenis Angelis, Evangelos Boukas
Abstract
More than 70% of the people that fell overboard cruise ships in the period 2010-2019 lost their lives. This paper presents a strategy for reliably predicting the area a person may be in after a man overboard incident, and describes in detail the search methods to find them utilizing UAV technology. The search area prediction method employs an Extended Kalman Filter that capitalizes on the information from the Leeway model to track the missing person in the sea by taking into account the uncertainty of the movement of the person and the weather conditions in the area. Then, a UAV uses this information to search for the person. Five different methods for searching in this dynamic area are presented and evaluated - the Zigzag, the Boustrophedon, the Spiral, the Probability Informed Search and the Improved Probability Informed Search (IPIS) methods. The IPIS method provides success rate of over 80% on average for finding a person, even if the UAV initiates the search mission 20 minutes after the man overboard incident and even assuming a detection method with a success rate of 30%. All code for the simulation environment and the evaluation of the methods is available on our GitHub page at https://github.com/diangeli/pdms-man-overboard.