Reconstruction of maritime route network in nearshore waters with noise removal using step-by-step density-based clustering

The increasing volume of marine traffic has significantly heightened the risk of accidents in nearshore and confined waters. While previous studies have primarily focused on large-scale maritime route networks, they often overlook the detailed delineation of ship channel profiles and centerlines. In high-risk areas, accurately mapping ship channel outlines and primary routes is crucial for traffic safety and route selection. Thus, this paper proposes a method for modeling maritime route networks in high-risk areas using Automatic Identification System data and Step-by-step Density-Based Spatial Clustering of Applications with Noise (S2-DBSCAN). First, stay trajectories are cleaned, and anomalies are identified. Then, three trajectory similarity metrics Hausdorff distance, Course Over Ground (COG) similarity of key turning points, and average COG similarity are used to design the S2-DBSCAN algorithm for trajectory clustering. Finally, the centerline of clusters is extracted using constrained Delaunay triangulation to form the route network. Experimental results on real data demonstrate that our method effectively removes noise and extracts the route network from ship trajectory clusters. In particular, the identified directional routes accurately represent ships’ true positions and movement patterns in geospatial space. This enhances route selection in busy waters, supports risk avoidance strategies, and aids maritime regulation.

日益增长的海上交通流量显著提升了近岸与受限水域的事故风险。尽管过往研究多聚焦于大规模海上航线网络，但往往忽略了船舶航道剖面与中心线的精细化划定。在高风险水域，精准绘制船舶航道轮廓与主要航线，对于交通安全与航线选择至关重要。据此，本文提出一种结合自动识别系统(Automatic Identification System, AIS)数据与分步密度聚类算法(Step-by-step Density-Based Spatial Clustering of Applications with Noise, S2-DBSCAN)的高风险水域海上航线网络建模方法。首先，对航行轨迹进行数据清洗并识别异常轨迹；随后，结合三种轨迹相似度指标——豪斯多夫距离(Hausdorff distance)、关键转向点的对地航向(Course Over Ground, COG)相似度以及平均对地航向相似度，设计用于轨迹聚类的S2-DBSCAN算法；最后，通过约束德劳内三角剖分(constrained Delaunay triangulation)提取聚类簇的中心线，以此构建航线网络。基于真实数据的实验结果表明，所提方法可有效剔除噪声并从船舶轨迹聚类结果中提取航线网络。尤为关键的是，识别出的定向航线能够精准反映船舶在地理空间中的真实位置与运动模式。该方法可优化繁忙水域的航线选择，辅助风险规避策略制定，并为海事监管提供支持。