Pattern-driven analysis of pedestrian movement

Abstract

Pattern mining is the most prominent topic in data mining. Many methods have been proposed to mine patterns, and clustering is one of the most popular methods. Clustering is the grouping of similar data items together. Numerous similarity measures have been proposed to determine the similarity between trajectories for clustering. As indoor Location Based Services (LBS) are maturing now, it is possible to fully track and record indoor movement trajectories, which was not possible until recently. However, it is still unclear which trajectory similarity measures are also effective for indoor environments.In this study, various similarity measures for trajectory clustering are studied to assess their efficacy for indoor pattern mining, and their performance is evaluated by the Silhouette Coefficient. Additionally, a framework for indoor pattern mining is proposed, emphasizing the semantic and spatial aspects of the trajectories. In the proposed framework, semantic patterns are mined first, followed by clustering of spatially similar trajectories participating in a semantic pattern.The results show that the Edit Distance-based metric distance measure, i.e., Edit Distance with Real Penalty (ERP), is more efficient. Furthermore, three out of four unknown venues were successfully predicted, which proves that the proposed framework is effective and a combination of semantic and spatial aspects of trajectories is crucial for indoor trajectory pattern mining, while the temporal aspect could provide added value. Therefore, in the future, it could be a valuable addition to the framework for indoor pattern mining.