基于时空聚类算法的轨迹停驻点识别研究

利用手机等智能移动终端获取个体出行GPS轨迹数据，提出了一种新的时空聚类算法AT-DBSCAN识别轨迹中的停驻点.该方法以固定长度的滑窗搜索核心点，以时空邻近条件定义簇间距离，以簇密度大小规定合并次序.提出了出行次数一致性、出行起止时刻误差、停驻点时长误差、停驻点中心偏移距离4个算法验证指标，弥补了传统查全率、查准率等忽略停驻点时空信息准确性的不足.结果表明，识别的停驻点有98%的位置误差在30m以内，100%的时长误差在5min以内，98%的出行起止时刻误差在5min以内.此外，算法对于室内活动、定位飘移、路径重合程度高等复杂轨迹具有较好的泛化能力.

Anchors Identification in Trajectory Based on Temporospatial Clustering Algorithm

Based on the individual GPS data collected by the mobile terminal like smartphones, a novel spatiotemporal clustering algorithm called AT-DBSCAN is proposed to identify anchors in the trajectories. In this algorithm, the core points are found and checked within a sliding window, the similarity is defined by the temporospatial difference, and clusters are merged according to the density. Rather than using the recall or precision to evaluate results, the consistency of travel times, the comprehensive bias when the travel starts and ends, the difference between anchors duration, and the central distance between the true and identified anchors, are proposed and applied to get the exact temporal-spatial information. Results suggest that 98 percent anchors’ distances are less than 30 meters, 100 percent duration difference are less than 5 minutes, and 98 percent travel time difference are within 5 minutes. Besides, the proposed algorithm also has a good performance in generalization of indoor activity, positioning drift and severe overlying segments.