A novel ship trajectory clustering method for Finding Overall and Local Features of Ship Trajectories

Abstract

Ship trajectory clustering is one of the main methods of ship trajectory mining based on AIS data. However, there exist two main problems in trajectory clustering: One is the inherent problem of the clustering algorithm, such as low computational efficiency, poor recognition of noise, sensitivity to the density distribution of data, etc.; the other problem is about the similarity measurement of the trajectories, that is, most of the measurement methods ignore the overall motion trend or local features of ship trajectories. In order to solve these problems, this paper proposes the FOLFST, a novel ship trajectory clustering method for Finding Overall and Local Features of Ship Trajectories, which consists of four parts: 1) segment the trajectory with the ADP (Adaptive-threshold Douglas-Peucker) algorithm; 2) adopt new sub-trajectory segments similarity measurement method which considers both the wholeness and locality of the trajectory; 3) cluster the sub-trajectory segments with the improved FOP-OPTICS (Finding of the Ordering Peaks Based on OPTICS); 4) identify sub-trajectories which belong to multiple clusters. The experimental results show that the FOLFST has obvious advantages over the other four algorithms in identifying the overall and local features of ship trajectories, noise recognition, computational efficiency, etc.