基于矢量场聚类的异常时空轨迹检测

发布时间：2018-02-01 03:32

本文关键词： 矢量场层次聚类加权异常检测　出处：《昆明理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：轨迹分析指的是对运动目标运行轨迹进行分析,以便获取运动目标的行为。轨迹异常检测就是通过对轨迹进行分析检测出其中出现的目标异常行为、异常事件。轨迹异常检测可应用于飓风、动物迁徙预测,交通流监测等方面。随着卫星定位数据、交通监控视频数据量的迅速增长,轨迹数据量与其包含的时空信息也迅速增长,然而通过人工分析数据的方式耗时耗力,且容易出现错误。本文利用聚类的方式将时空轨迹数据划分为不同的簇,通过计算聚类中心轨迹与待检测轨迹之间的距离从而自动判别时空轨迹正常与否,以便有效解决各类时空数据分析应用。本文首先简要分析了从视频数据中获取运动目标轨迹的几种常见方法的优缺点。其次,提出一种矢量场层次聚类的方法对轨迹数据进行聚类,解决矢量场轨迹聚类不能自适应聚类类别数的问题,并且通过加权矢量场拟合解决噪声轨迹点对聚类结果的干扰,增强了算法的鲁棒性。最后,通过计算检测数据矢量场与各聚类中心轨迹矢量场的相似度,判定待测试轨迹正常与否。通过对监控视频数据上进行的实验表明,本文提出的轨迹聚类方法与传统的轨迹聚类相比具有更高的类别适应性与鲁棒性,对异常轨迹检出率达到90%以上。
[Abstract]:Trajectory analysis refers to the trajectory analysis of the moving target, in order to obtain the target behavior. Anomaly detection is based on the trajectory analysis to detect abnormal behavior, which targets abnormal events. Trajectory trajectory outlier detection can be applied to the hurricane, animal migration prediction, traffic flow monitoring. With the development of satellite positioning data the rapid growth of traffic monitoring, the amount of video data, and contains temporal information track data are also growing rapidly. However, through artificial way of analyzing data is time-consuming, error and easy to use. In this way the clustering of trajectory data into different clusters, by calculating the cluster center trajectory and trajectory to be detected between the distance to automatically determine the spatio-temporal trajectory is normal or not, in order to effectively solve various spatio-temporal data analysis applications. This paper briefly analyzes from the optic frequency According to the advantages and disadvantages of several common methods for moving target. Secondly, put forward a kind of vector field hierarchical clustering method to cluster the trajectory data, solve trajectory clustering is not adaptive clustering number vector field, and by weighted vector field fitting to solve noise track points on the clustering results, robustness the algorithm. Finally, by calculating the detection data of vector field and the cluster center trajectory vector field test trajectory similarity, determined to be normal or not. Based on the video surveillance data. Experimental results show that, compared with the trajectory clustering trajectory clustering method proposed in this paper with the traditional categories of adaptability and greater robustness, the abnormal the trajectory detection rate reached more than 90%.

【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X924.2;TP391.41

【参考文献】