基于轨迹特征的船舶停留行为识别与分类

为准确评估大规模轨迹数据中的船舶停留活动，构建了两阶段船舶轨迹停留点提取策略，提出了特征驱动的船舶停留行为识别与自动分类方法；以距离、时间和轨迹点数量为约束条件构建了规则模型，检测了原始轨迹中的停留候选轨迹，引入孤立森林算法检测和去除异常离群点，提取了高聚集度的船舶停留轨迹集合；基于船舶靠泊和锚泊的时空特征，定义了轨迹点重复率、相邻点平均距离和最远点对距离3个指标，构建了新的轨迹相似性度量模型，量化了船舶停留轨迹点的分布特征和聚合程度，并利用K近邻算法完成了船舶锚泊行为与靠泊行为的自动分类；采用提出的方法处理了3个不同水域的船舶轨迹数据，准确获取了船舶停留行为的分类结果，并验证了船舶锚泊与靠泊在轨迹时空特征上的差异性，以人工标注结果为参考依据评估了船舶停留行为识别与分类的准确性。研究结果表明:船舶靠泊的轨迹点重复率在80%以上，最远点对距离和相邻点平均距离分别为6~11和1~2 m，船舶锚泊的轨迹点重复率在10%以下，最远点对距离和相邻点平均距离分别为150~250和8~10 m，说明轨迹点重复率、相邻点平均距离和最远点对距离这3个时空特征对船舶靠泊和锚泊具有显著的区分能力；提出的方法对船舶停留识别分类的正确率在98%以上，充分证明了其有效性；采用提出的方法可更新已有码头和锚地的空间位置，自动识别规则水域外的船舶异常停留和规则水域内的超长时间船舶异常停留，掌握在港船舶停留分布情况，识别不同季节、不同时段的热点码头和锚地，从而辅助优化港口规划布局和交通组织。 

Stopping behavior recognition and classification of ship based on trajectory characteristics

To estimate stopping activities of ships from massive trajectory data accurately, a two-stage strategy was established to extract stop points from ship trajectories, and an automatic characteristic-based ship stopping behavior recognition and classification method was also proposed. By taking the distance, time and number of points as the constraint conditions, a rule model was constructed to detect the candidate stop trajectories from the raw trajectories. The isolation forest algorithm was applied for the abnormal outliers detection and elimination. A set of highly clustered ship stop trajectories was extracted. Based on the spatio-temporal characteristics of ship berthing and anchoring. Three indices, including the repetition rate of trajectory point, mean distance between neighboring points, and distance between the farthest point pair, were defined to establish a new trajectory similarity measurement model. Then, the distribution characteristics and aggregation degree of ship stop trajectory points were quantitatively evaluated, and the K-nearest neighbor algorithm was then used to automatically classify the berthing and anchoring behaviors of ships. The proposed method was applied to the ship trajectory data collected from three different waters. The classification results of ship stopping behaviors were obtained accurately. The differences in spatio-temporal characteristics of ship anchoring and berthing were verified. The accuracies of recognition and classification of ship stopping behaviors were assessed with the help of manually annotated results. Research results indicate that the repetition rate of trajectory points for ship berthing is more than 80%. The distance between the furthest point pair and the mean distance between neighboring points are 6-11 and 1-2 m, respectively. The repetition rate of trajectory points for ship anchoring is less than 10%. The distance between the furthest point pair and the mean distance between neighboring points are 150-250 and 8-10 m, respectively. Thus, the three spatio-temporal characteristics, including the repetition rate of trajectory point, mean distance between neighboring points, and distance between the farthest point pair have a significant ability to distinguish the ship berthing and anchoring. The recognition and classification accuracy of the proposed method reaches up to 98%. Therefore, its effectiveness is fully proved. With the help of the proposed model, the spatial positions of existing docks and anchorages can be updated. Abnormal ship stops outside the regular waters or abnormal ship stops for prolonged periods inside the regular waters can be recognized automatically. The stopping distribution in ports can be monitored, and the popular docks and anchorages in different times and seasons can be known. In this way, the port planning layout and traffic organization can be optimized. 3 tabs, 7 figs, 31 refs.