Ship Trajectory Prediction with Social History LSTM

Ship trajectory prediction is an important research topic in ship navigation automation. It can effectively help ship pilots to obtain comprehensive Marine traffic information and reduce the potential collision risk of ships. Compared with pedestrian and vehicle trajectory prediction, due to the existence of AIS system for ships, it is easier to obtain historical data of ships, so it is easier to explore the periodicity of trajectories in time. Therefore, based on the previous research on trajectory prediction, we propose a new history module, which adds the agent's long-term route intention to the model to guide the model to predict the ship trajectory. At the same time, we adopt a data-driven idea and pre-train a State Refinement module to obtain high-dimensional feature representations of velocity. Experiments show that our model is superior to the existing current algorithms in long-term trajectory prediction on self-made data sets.