IEEE Trans Vis Comput Graph. 2018 Jan;24(1):34-44. doi: 10.1109/TVCG.2017.2744322. Epub 2017 Aug 29.
Clustering of trajectories of moving objects by similarity is an important technique in movement analysis. Existing distance functions assess the similarity between trajectories based on properties of the trajectory points or segments. The properties may include the spatial positions, times, and thematic attributes. There may be a need to focus the analysis on certain parts of trajectories, i.e., points and segments that have particular properties. According to the analysis focus, the analyst may need to cluster trajectories by similarity of their relevant parts only. Throughout the analysis process, the focus may change, and different parts of trajectories may become relevant. We propose an analytical workflow in which interactive filtering tools are used to attach relevance flags to elements of trajectories, clustering is done using a distance function that ignores irrelevant elements, and the resulting clusters are summarized for further analysis. We demonstrate how this workflow can be useful for different analysis tasks in three case studies with real data from the domain of air traffic. We propose a suite of generic techniques and visualization guidelines to support movement data analysis by means of relevance-aware trajectory clustering.
通过相似性对移动对象的轨迹进行聚类是运动分析中的一项重要技术。现有的距离函数基于轨迹点或段的属性来评估轨迹之间的相似性。这些属性可以包括空间位置、时间和主题属性。可能需要将分析重点放在轨迹的某些部分上，即具有特定属性的点和段。根据分析重点，分析师可能需要仅根据相关部分的相似性对轨迹进行聚类。在整个分析过程中，重点可能会发生变化，轨迹的不同部分可能会变得相关。我们提出了一种分析工作流程，其中使用交互式过滤工具将相关性标志附加到轨迹的元素上，使用忽略不相关元素的距离函数进行聚类，并对生成的聚类进行总结以供进一步分析。我们通过三个案例研究展示了该工作流程如何在航空交通领域的真实数据中对不同的分析任务有用。我们提出了一套通用技术和可视化准则，通过基于相关性的轨迹聚类来支持运动数据分析。
