A Novel Machine-Learning Framework With a Moving Platform for Maritime Drift Calculations

Abstract

A novel data-driven conceptual framework using a moving platform was developed to accurately estimate the drift of objects in the marine environment in real time using a combination of a perception-based sensing technology and deep-learning algorithms. The framework for conducting field experiments to establish the drift properties of moving objects is described. The objective of this study was to develop and test the integrated technology and determine the leeway drift characteristics of a full-scale three-dimensional mannequin resembling a person in water (PIW) and of a rectangular pelican box to accurately forecast the trajectory and the drift characteristics of the moving objects in real time. The wind and ocean current speeds were measured locally for the entire duration of the tests. A sensor hardware platform with a light detector and ranging sensor (LiDAR), stereoscopic depth cameras, a global positioning system, an inertial measurement unit, and operating software was designed and constructed by the team. It was then mounted on a boat (mobile test platform) to collect data. Tests were conducted by deploying the drifting objects from the mobile test platform into Galveston Bay and tracking them in real time. A framework was developed for applying machine learning and localization concepts on the data obtained from the sensors to determine the leeway trajectory, drift velocity, and leeway coefficients of the drifting objects in real time. Consistent trends in the downwind and crosswind leeway drift coefficients were observed for the pelican (significantly influenced by the wind) and PIW (influenced by the winds and currents).