Joint Stochastic Prediction of Vessel Kinematics and Destination based on a Maritime Traffic Graph

Maritime traffic increases every year, and therefore also the amount of traffic data from the Automatic Identification System (AIS). Utilizing these data on traffic patterns and possible destinations for long-term vessel prediction is here an important way of gaining maritime situational awareness (MSA) for use in the decision making in autonomous ships, such as collision avoidance algorithms, which can reduce collision risk during voyage. In this article, we present a destination inference method based on piece-wise Ornstein-Uhlenbeck (OU) processes for predicting vessel motions along common traffic lanes toward a set of destinations. The mean velocities of the processes are inferred through the creation of a maritime graph that represents the major traffic patterns in the area of consideration. After using an OU process for prediction along major sea lanes, the Equilibrium Reverting Velocity (ERV) bridging model is used to enable convergence of the prediction towards considered destinations. Then, based on the OU process and ERV bridging model for prediction, Bayesian inference is used to estimate the posterior destination distribution. Tested on a real-time AIS dataset, the method is shown to perform better than current state-of-the art methods in destination inference as it indirectly takes land and passed destinations into account.