Computer Engineering and Applications ›› 2018, Vol. 54 ›› Issue (1): 77-85.DOI: 10.3778/j.issn.1002-8331.1612-0528

Previous Articles     Next Articles

HMM-Cluster: Trajectory clustering for discovering traffic volume overload

PAN Li1,2, DENG Jia1, WANG Yongli1   

  1. 1. School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China
    2. Staff of PLA Rocket Force, Beijing 100085, China
  • Online:2018-01-01 Published:2018-01-15


潘  立1,2,邓  佳1,王永利1   

  1. 1.南京理工大学 计算机科学与工程学院,南京 210094
    2.中国人民解放军火箭军参谋部,北京 100085

Abstract: With the development of economy, the urban traffic congestion has become an urgent problem in China. The traffic volume overload discovering is an effective method for solving the problem of traffic congestion. A kind of trajectory clustering method based on the HMM model, named HMM-Cluster, is put forward, which can find out the traffic volume overload conditions. HMM-Cluster extracts the feature points of spatio-temporal trajectory data firstly, and it uses dimension reduction technique to decrease the trajectory data volume, as well as save the cost of storage. Secondly, it trains a HMM model for each reference trajectory based on density function to get a trajectory affinity similarity matrix. Finally, the HMM-Cluster algorithm aggregates similarity trajectory effectively and forms the clustering results of trajectory data. The contrast experiments on actual data prove that the HMM-Cluster method has a good effect, which can obtain moving objects’ pattern and discover traffic volume overload effectively and conveniently. The proposed method has significant values in real application.

Key words: traffic volume overload, spatio-temporal data, trajectory clustering, Hidden Markov Model(HMM)

摘要: 随着经济的发展,城市交通拥堵问题亟待解决,交通量过载发现是解决交通拥堵问题的有效方法之一。提出一种基于HMM模型的轨迹聚类算法HMM-Cluster,可有效地发现交通量过载情况。该算法首先提取时空轨迹特征点,并采用维数约简技术减少轨迹数据量,根据参照轨迹拟合HMM模型,基于密度函数得到轨迹相似度矩阵,最后给出聚合的相似性轨迹。真实轨迹数据集上的对比实验结果表明,提出的HMM-Cluster可有效地挖掘移动对象运动模式,准确发现交通量过载情况,具有一定实用价值。

关键词: 交通量过载, 时空数据, 轨迹聚类, 隐马尔科夫模型