基于多维航迹特征的异常行为检测方法  

Anomalous behavior detection method based on multidimensional trajectory characteristics

作　　者：潘新龙[1] 王海鹏[1] 何友[1] 熊伟[1] 周伟[1] 

机构地区：[1]海军航空工程学院信息融合研究所,烟台264001

出　　处：《航空学报》2017年第4期249-258,共10页Acta Aeronautica et Astronautica Sinica

基　　金：国家自然科学基金(61531020;61471383;91538201);山东省科技重大专项基金(2015ZDZX01001)~~

摘　　要：在信息融合领域,利用数据挖掘中的异常检测技术,可以基于目标的多维航迹特征来挖掘目标的异常行为。现有轨迹异常检测方法主要检测目标的位置异常,没有充分利用目标的属性、类型、位置、速度和航向等多维特征,在挖掘目标的异常行为时具有局限性。通过定义多因素定向Hausdorff距离和构造多维度局部异常因子,提出了一种基于多维航迹特征的异常行为检测方法,通过对多维航迹数据的异常检测,实现对目标异常行为的挖掘。在仿真军事场景和真实的民用场景上进行了实验分析,所提方法都能有效的检测出目标的异常行为。In the information fusion domain, anomalous behaviors could be mined based on multidimensional trajectory characteristics by using the anomalous detection technique in data mining. Previous trajectory anomaly detection algorithms mainly detect the position anomalies, without making full use of the attribute, category, position, velocity, and course characteristics. In order to overcome this limitation, we define the multi-factor Hausdorff distance, construct the multidimensional local outlier factor, and propose a method for detecting anomalous behaviors based on multidimensional trajectory characteristics. The method can mine anomalous behaviors based on detecting multidimensional trajectories. We conducted experiments on simulated military scenario and real civilian scenario, the proposed method can effectively detect the anomalous behavior of the target.

关 键 词：异常行为 航迹 多维特征 局部异常因子 HAUSDORFF距离 

分 类 号：V355.1[航空宇航科学与技术—人机与环境工程] TP274.2[自动化与计算机技术—检测技术与自动化装置]