基于聚类的出租车异常轨迹检测

Clustering-based Taxi Trajectory Outlier Detection

出租车全球定位系统数据中蕴含城市交通和移动对象行为的宏观信息,从中可以挖掘出有价值的异常轨迹模式。将位置和几何形状、行驶时间分别作为出租车轨迹的空间与时间特征,根据特征偏离情况划分时间、空间和时空异常轨迹。从轨迹数据中提取相同起终点的轨迹集,将轨迹划分成轨迹片段,计算轨迹间的相似度并进行基于距离和密度的聚类,在空间特征上初步分离出频繁和稀疏轨迹,根据数据异常判定的kσ准则确定时间特征异常的分离阈值,对时间特征进行再次划分,最终实现出租车异常轨迹检测。实验结果表明,该方法能从异常轨迹中挖掘出个性化路线、异常停留位置和交通路段,为智能交通、物流高效规划和执行等提供参考信息。

Taxi Global Position System（GPS） data contain macro information about the behavior of urban traffic and moving object behavior,from which valuable anomalous trajectory patterns can be mined.The location,geometry and travel time are taken as the spatial and temporal characteristics of the taxi trajectory respectively.According to the deviation of the feature,the trajectory anomalies are divided into temporal,space and spatio-temporal outliers.The trajectories of the same starting and ending points are extracted from the trajectory data,and are partitioned into segments.The similarity between trajectories is calculated and clustering based on distance and density is carried out.Frequent and the sparse trajectories are preliminary separated by the spatial characteristies.Based on κσ criterion,the separation threshold of temporal anomaly is determined to realize the classification of the temporal characteristic,and finally the trajectory outlier detection of the taxi is realized.The experimental results show that the method can extract personalized route as well as abnormal parking location and traffic section from abnormal trajectories providing reference information for intelligent transportation as well as efficient logistics planning and execution.