Man overboard event detection from RGB and thermal imagery

Abstract

A man overboard is an emergency incident, in which fast detection is the most crucial factor, for the quickest and most efficient recovery of the victim. As such, efficient monitoring methodologies should be employed. A variety of sensors is available today, supporting a continuous monitoring process, regardless of environmental conditions; RGB and thermal are two commonly used sensors. At the same time, several algorithms and techniques have been tested and proved to be efficient in human detection and situation recognition tasks. However, to this day a coherent methodology for fall detection over multiple sensors on a large-scale deployment, complying with related ISO standards on extremely low false positive alerts, has not been implemented. In this paper, we investigate the possibilities as well as the limitations of man overboard vision-based systems' development based on RGB and thermal imagery.