中文摘要：

轨迹停留蕴含重要语义信息,其有效提取是开展轨迹Stop/Move模型分析的前提。本文首先依据核密度思想,通过累计邻域点时空贡献来定义轨迹点的停留指数,在此基础上设计了停留指数图,以图形方式直观表达轨迹点的时空聚集程度变化。进一步针对源于停留指数的潜在停留段,提出了一种基于潜在停留段时空临近关系的逐级合并算法,以自动发现和提取停留。试验表明,该算法兼顾停留识别的完整性和准确性,可以有效识别复杂多样的轨迹停留,即使面对噪声严重的轨迹,停留提取的正确率依然较高。

英文摘要：

Trajectory stops imply important semantic information,and the extraction of trajectory stops is the premise to carry out advanced Stop/Move analysis. This paper,based on the idea of kernel density,firstly introduces the concept of stop index,which is derived by cumulating spatio-temporal contribution of neighboring points,and further designs stop index graph to intuitively visualize the evolution of spatio-clustering degree during a trajectory. Importantly,stops index and its graph are related to spatial scale through neighboring radius,which then can be exploited to analyze trajectory stops under multiple scales. In addition,this paper introduces stop sequence rooted from stop index,and proposes an algorithm for the automatic extraction of trajectory stops by progressively merging stop sequences. According to the algorithm,a stop under strong GPS signal exactly corresponds to a stop sequence,while a stop under weak GPS signal could be derived by merging multiple stop sequences. Experiments based on own-acquired and Geo Life trajectories show that the algorithm has achieves the balance between the completeness and accuracy of stop extraction,and could effectively discover and extract complex and diverse trajectory stops. Even facing trajectories with serious drift noises,the algorithm still achieves a high rate of accuracy on stop extraction.