摘要： 目前的轨迹数据聚类直接对整条轨迹数据聚类，或先分段再对轨迹段聚类。分段聚类法抛弃轨迹段内部点，丢失轨迹局部特征，没有考虑点的速度影响。针对该问题，提出一种基于速度约束的分段轨迹数据聚类方法。该方法将速度约束和two-pass corner detection应用于轨迹分段，在邻域计算中加入速度约束，采用Discrete Fréchet Distance比较轨迹段距离，保留了轨迹段内部特征。用类似DBSCAN的TraDBSCAN算法对轨迹段进行聚类。实验结果表明，该方法考虑速度因素，可以获得更有效的聚类结果。

关键词: 轨迹, 分段轨迹, 离散弗雷歇距离, TraDBSCAN算法, 轨迹聚类

Abstract: The current trajectory data clustering algorithms do clustering directly on the whole trajectory, or on sub-trajectory after partitioning. The sub-trajectory clustering algorithms abandon all the points within the sub-trajectory, that is to say, this kind of algorithms loose sub-trajectory’s internal features, while not considering the speed impact of points. A method for sub-trajectory data clustering taking the speed impact on consideration is proposed. The method uses speed restriction and two-pass corner detection in the segment of trajectories. It adds speed restriction in neighborhood computations and do sub-trajectory distance compared with Discrete Fréchet Distance, which keeps the internal features of sub-trajectory. Besides, TraDBSCAN algorithm that is similar to DBSCAN algorithm is applied for the clustering of sub-trajectory. Experimental results show that the new algorithm is effective and it perfectly considers the speed factor.

Key words: trajectory, sub-trajectory, discrete Freshet distance, TraDBSCAN algorithm, trajectory clustering

中图分类号: 

• TP18