基于关键特征点的航迹高效聚类方法  

Efficient trajectory clustering method based on key features

作　　者：王建[1] 王军 宋巍 黄冬梅[1,2] 闫丹凤 郑小罗 WANG Jian;WANG Jun;SONG Wei;HUANG Dongmei;YAN Danfeng;ZHENG Xiaoluo(College of Information Technology,Shanghai Ocean University,Shanghai 201306,China;Shanghai University of Electric Power,Shanghai 201306,China)

机构地区：[1]上海海洋大学信息学院,上海201306 [2]上海电力大学,上海201306

出　　处：《海洋测绘》2023年第2期35-39,共5页Hydrographic Surveying and Charting

基　　金：上海市科委部分地方高校能力建设项目(20050501900);上海市科委地方院校能力建设项目(20020500700)。

摘　　要：针对传统的聚类方法存在计算量大、度量距离定义难的问题,提出了一种基于关键特征点的航迹高效聚类方法。首先,为降低聚类时的计算量,对用于原始航迹压缩的D-P算法进行了改进,实现了对局部短航迹的去除和航迹关键特征点的保留;然后,根据船舶“位置”和“航向”定义了子航迹间的相似性度量,并结合DTW算法思想提出了一种基于子航迹的度量距离DBLD。实验表明,经改进的D-P算法压缩处理后,航迹间度量距离的计算量降低了29.27%;在3种经典聚类算法中,使用DBLD距离的聚类效果优于融合距离、Hausdorff距离。该方法可为进一步研究航迹分布提供有益参考。To solve the problems of large computation and difficult definition of metric distance in traditional clustering methods,an efficient trajectories clustering method based on key feature points is proposed.Firstly,in order to reduce the calculated amount of clustering,the D-P algorithm used for trajectories compression is improved to remove the local short trajectories and retain the key feature points of the raw trajectories.Then,the similarity measurement between sub-trajectories is defined according to the"location"and"direction",and a metric distance DBLD(Distance Based on Location and Direction)is proposed combining with the idea of DTW algorithm.Experimental results show that,after being compressed by the improved D-P algorithm,the calculation amount of metric distance between trajectories is reduced by 29.27%.Among the three classical clustering algorithms,DBLD distance,fusion distance and Hausdorff distance are used to compare the clustering effect.The clustering effect based on DBLD distance is better than the other two metric distances.The method presented in this paper provides a useful theoretical reference for further research on trajectories distribution.

关 键 词：航海数据分析 航迹聚类 相似性度量 航迹压缩 AIS数据 

分 类 号：P229[天文地球—大地测量学与测量工程]