基于加权矢量场的轨迹层次聚类

Trajectory hierarchical clustering based on weighted vector field

传统的轨迹聚类方法存在定义轨迹相似度难度大,聚类过程中容易忽略轨迹细节等问题。基于矢量场的轨迹聚类(VFC)在保持轨迹原始运动特征的基础上,利用矢量场的几何结构可以很好地度量轨迹相似度。引入加权拟合方法,降低噪声对聚类的影响,以解决VFC鲁棒性较差问题。采用层次聚类动态地决定聚类类别数,以解决聚类类别数不能自适应的问题,提高聚类有效性。采用亚特兰大飓风数据作为实验原始轨迹数据,分别使用经典矢量场的轨迹聚类,k-means聚类,k-mediods聚类以及提出的方法进行实验,实验结果证明了加权拟合矢量场的层次聚类算法的有效性。

It is hard to define similarity of trajectories and trends to ignore details of trajectories using a traditional trajectory clustering method. Vector field based clustering methods keep the inherent features of movements and measures similarities of trajectories with the geometric structure information derived from it. Weighted fitting scheme is introduced to weaken the effects of noises and increase the robustness of clustering. A hierarchical approach is employed to automatically determine the number of class solving the problem that traditional methods cannot be self-adaptive to clustering,thus improving the effectiveness of our method. Experiments of traditional vector field clustering,k-means clustering and k-mediods clustering as well as the proposed method are conducted on the Atlanta Hurricane dataset,and the result shows the effectiveness of the hierarchical clustering algorithm based on weighted vector field.