An AIS-Based Multiple Trajectory Prediction Approach for Collision Avoidance in Future Vessels

Abstract

This study presents a method to predict the future trajectory of a target vessel using historical AIS data. The purpose of such a prediction is to aid in collision avoidance in future vessels. The method presented in this study extracts all trajectories present in an initial cluster centered about a vessel position. Features for each trajectory are then generated using Principle Component Analysis and used in clustering via unsupervised Gaussian mixture modeling. Each resultant cluster represents a possible future route the vessel may follow. A trajectory prediction is then conducted with respect to each cluster of trajectories discovered. This results in a prediction of multiple possible trajectories. The results indicate that the algorithm is effective in clustering the trajectories, where at least one cluster corresponds to the true trajectory of the vessel. The resultant predicted trajectories are also found to be reasonably accurate.