基于改进Hausdorff距离的轨迹聚类算法  

Trajectory Clustering Algorithm Based on Improved Hausdorff Distance

作　　者：陈锦阳[1,2] 宋加涛[1] 刘良旭[1] 王让定[2] 

机构地区：[1]宁波工程学院电子与信息工程学院,浙江宁波315016 [2]宁波大学信息科学与工程学院,浙江宁波315211

出　　处：《计算机工程》2012年第17期157-161,共5页Computer Engineering

基　　金：国家自然科学基金资助项目(60972163);浙江省自然科学基金资助项目(Y1100598);信息处理与自动化技术浙江省重中之重学科开放基金资助项目(201100808);浙江省综合信息网技术重点实验室开放基金资助项目(201109);宁波市自然科学基金资助项目(2009A610090;2011A610175)

摘　　要：以整条轨迹为目标的聚类方法存在轨迹较长的问题。为此,提出一种以轨迹子段为聚类目标的聚类算法CTIHD。给出一种新的轨迹子段距离度量方法,用以消除轨迹子段之间的公共偏差。利用特征点概念将轨迹划分成轨迹子段集,计算轨迹子段之间的相似度,由此实现聚类。实验结果表明,该算法相比同类算法具有更好的轨迹聚类效果。For problems which the whole trajectory as the target for the clustering, this paper proposes a clustering algorithm called CTIttD（Cluslering of Trajectories based on hnproved ttausdorff Distance）, which uses a sub-trajectory as the target for the clustering. In this algorithm, in order to effectively calculate the similarity between the trajectory, the algorithm defines a new sub-trajectory distance metrics, the definition can not only effectively eliminate the public error between sub-trajectory, but also take full account of sub-trajectory contains the movement featnre. In algorithm, trajectory is divided into sub-trajectories uses the concept of the trajectory of feature point,It uses the proposed the definition of trajectory distance metrics between sub-trajectories to calculated similarity between sub-trajectories; On this basis the use of traditional clustering methods for sub-trajectory clustering. Experimental results show that the algorithm can achieve better trajectory clustering effect than the existing methods.

关 键 词：轨迹聚类 运动模式 HAUSDORFF距离 点特征矩阵 轨迹子段 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]