基于子空间聚类算法的时空轨迹聚类

Spatio-temporal Trajectory Clustering Based on Automatic Subspace Clustering Algorithm

已有的时空轨迹聚类方法一般以整条轨迹作为聚类单元,聚类效果较低且不能识别轨迹局部特征;另一种轨迹聚类方法是以划分后轨迹段为聚类单元,算法效率较低且不能很好地支持多属性聚类。该文提出基于子空间聚类算法的时空轨迹聚类。首先引入数据归约的思想,将轨迹进行离散化处理,再运用CLIQUE算法对离散化后的轨迹段进行聚类。实验结果表明,此轨迹聚类方法具有较高的伸缩性,能有效地处理多维轨迹数据并识别轨迹的局部聚类特征,能揭示时空轨迹在不同子空间的运动规律。

For current trajectory clustering algorithm,some of them take the whole trajectory as clustering units.But these algorithms are low in efficient and are not able to identify the local feature of trajectories.Another kind of trajectory clustering algorithms firstly divide the whole trajectory into sub-trajectory,then cluster the sub-trajectory as basic units.These algorithms are low in efficient and are not able to deal with multi-attribute trajectory data.In this paper,in allusion to these weaknesses,a new way to partition trajectory and clustering sub-trajectory is presented.The algorithm firstly uses the idea of data reduction to discretize the whole trajectory,and then use CLIQUE algorithm to cluster the discretized trajectory.The result shows that the algorithm is high in scalability,and is able to cluster multi-attribute trajectory.The local clustering feature and motion regular pattern can be disclosed in various sub-space by the algorithm.