Abstract: For trajectory outlier detection problem of the taxi, on the basis of the existing taxi GPS data, and combined with the urban road intersection information, this paper puts forward an IBATD（Intersection-Based Anomalous Trajectories Detection） algorithm. The algorithm describes the GPS point in the form of intersection after map matching, and then clusters these trajectories with multiway-tree method. By calculating the trajectory probability under the test and comparing with the given anomaly threshold, it classifies the trajectory to be normal or abnormal. Compared with the classic spectral clustering algorithm based on Hausdorff distance, the multiway-tree clustering method has more accurate trajectory model library, faster operation speed, and can make real-time detection.

Key words: abnormal trajectory detection, Global Positioning System（GPS） traces, intersection information, multiway-tree clustering

摘要： 针对出租车的异常轨迹检测问题，根据已有的出租车GPS数据，结合城市道路路口信息，提出了一种基于路口的异常轨迹检测算法（Intersection-Based Anomalous Trajectories Detection，IBATD）。该算法将GPS数据进行地图匹配，并将匹配后的GPS轨迹以路口的形式描述，再以多叉树的方式实现轨迹聚类。通过计算待测轨迹的轨迹概率，并与给定异常阈值进行比较，将轨迹分类为正常或异常。与经典的基于Hausdorff距离的谱聚类算法相比，多叉树轨迹聚类具有更准确的轨迹模型库、更快的运算速度以及实时检测的特点。

关键词: 异常轨迹检测, GPS跟踪, 路口信息, 多叉树聚类