一种新的轨迹聚类及有效性评价方法

为解决轨迹聚类问题,提出一种新的无监督轨迹聚类及聚类有效性评估方法。通过建立双层字符串轨迹模型,计算得到轨迹间距离并用作聚类依据。提出轨迹同距点比例的概念,以此作为聚类工具,并采用类内平均同距点比例作为聚类有效性评价值。利用麻省理工大学(Massachusetts Institute of Technology,MIT)停车场行人路径数据集进行实验,实验结果表明,新的无监督聚类算法能较好地完成轨迹聚类任务,平均类内同距点比例能够很好地衡量分类效果。

A New Method of Trajectory Clustering and Validity Evaluation

In order to solve the problem of trajectory clustering, this paper proposes a new method of unsupervised trajectory clustering and validity evaluation. The clustering is based on the distance matrix calculated in the double-layer alphabetic string model. The paper also proposes the conception of PSD （Points With Same Distance） and regard it as the tool of trajectory clustering. The average proportion of PSD intra-clusters is used to evaluate the validity of the clustering. We test the method making use of the MIT （Massachusetts Institute of Technology ） parkinglot dataset. As the experiment shows, our unsupervised method can successfully accomplish the task of trajectory clustering and the proposed method of validity evaluation can express the validity of trajectory clustering.