航空学报 > 2023, Vol. 44 Issue (5): 326394-326394   doi: 10.7527/S1000-6893.2021.26394

时空轨迹多维特征融合的行为规律挖掘算法

姜乔文1, 刘瑜1(), 谭大宁1, 孙顺1, 董凯1,2   

  1. 1.海军航空大学 信息融合研究所,烟台 264000
    2.中国电子科学研究院,北京 100041
  • 收稿日期:2021-09-15 修回日期:2021-10-15 接受日期:2021-11-17 出版日期:2023-03-15 发布日期:2021-11-23
  • 通讯作者: 刘瑜 E-mail:liuyu77360132@126.com
  • 基金资助:
    国家自然科学基金(62022092);中国博士后科学基金(2020M680631)

Regular behaviors mining algorithm based on fusion of multidimensional features of spatiotemporal trajectory

Qiaowen JIANG1, Yu LIU1(), Daning TAN1, Shun SUN1, Kai DONG1,2   

  1. 1.Institute of Information Fusion,Naval Aviation University,Yantai 264000,China
    2.China Academy of Electronics and Information Technology,Beijing 100041,China
  • Received:2021-09-15 Revised:2021-10-15 Accepted:2021-11-17 Online:2023-03-15 Published:2021-11-23
  • Contact: Yu LIU E-mail:liuyu77360132@126.com
  • Supported by:
    National Natural Science Foundation of China(62022092);China Postdoctoral Science Foundation(2020M680631)

摘要:

在预警监视系统中,利用数据挖掘技术可以从海量的目标时空轨迹数据中挖掘出目标的行为规律,实现态势信息的智能感知。目前大部分行为规律挖掘方法仅考虑目标轨迹的空间位置信息,忽略了航向和速度信息,因此难以区分空间位置相似但运动速度和方向不同的行为。除此之外,轨迹聚类算法普遍存在参数设置复杂的问题,而且容易受到轨迹行为分布密度的影响。针对上述问题,首先,通过构造时间滑窗定义了时空Hausdorff距离,可度量时空轨迹多维特征差异;其次,结合k最近邻和密度峰值聚类中决策图的思想,提出了时空轨迹多维特征融合的行为规律挖掘算法;最后,使用仿真飞行器轨迹和实测雷达轨迹数据进行实验分析和验证,结果表明在典型应用场景下本文算法可以准确地挖掘出目标所有行为规律,在智能监视任务中具有较好的应用前景。

关键词: 行为规律挖掘, 时空轨迹, 多维特征, Hausdorff距离, 轨迹聚类

Abstract:

In the early warning and surveillance system, the data mining technology can be used to mine the regular behavior from the massive target trajectory data, and realize intelligent awareness of situation information. At present, most regular behaviors mining methods only consider the spatial position information of target trajectory, ignoring the course and speed information of the target. It is thus difficult to distinguish the behaviors with similar spatial positions but different speeds and directions. In addition, trajectory clustering algorithms generally have complex parameter settings, and are easy to be affected by the distribution density of trajectory behavior. To solve the above problems, the spatiotemporal Hausdorff distance is firstly defined by constructing a time sliding window, which could measure the difference of multidimensional features of spatiotemporal trajectory. Secondly, based on the idea of k-nearest neighbor and the decision graph in density peak clustering, a regular behaviors mining algorithm is proposed based on fusion of multidimensional features of spatiotemporal trajectory. Finally, the simulated aircraft trajectory data and measured radar trajectory data are used for experimental analysis and verification. The results show that the proposed algorithm can accurately mine all behavior laws of targets in typical application scenarios, and has possible applications in intelligent surveillance tasks.

Key words: regular behaviors mining, spatiotemporal trajectory, multidimensional features, Hausdorff distance, trajectory clustering

中图分类号: