A trajectory clustering method based on Douglas-Peucker compression and density for marine traffic pattern recognition

Clustering analysis is applied extensively in pattern recognition. In marine traffic applications, the clustering results may exhibit a customary route and traffic volume distribution. In order to improve the clustering performance of ship trajectory data, which is characterized by a large data volume and distribution complexity, a method consisting of Douglas-Peucker (DP)-based compression and density -based clustering is proposed. In the first part of the proposed method, the appropriate parameters for the DP algorithm were determined according to the shape changes in the trajectories, which were used to compress the trajectories prior to calculating the dynamic time warping (DTW) distance matrix. In the second part, the density -based spatial clustering of applications with noise (DBSCAN) algorithm was improved in terms of determining the parameters. Based on the statistical characteristics of ship trajectory distribution, the appropriate DBSCAN parameters could be determined adaptively. Evaluation and comparison experiments were conducted based on massive real ship trajectories in the Chinese port of Beilun-Zhoushan. The results demonstrated that, compared to the traditional DTW distance, the proposed similarity measurement exhibits superior performance in terms of both time and quality. Furthermore, the results of the comparison experiment demonstrated that the improved DBSCAN outperforms two existing clustering methods in marine traffic c pattern recognition.