Understanding Traffic Patterns using Clustered Semantic Trajectories and Local Geographic Units

Revealing traffic behavior from GPS data is a possibility to create a current and detailed data basis for city and traffic planning. Currently, traffic planning is mostly done by surveys or simulations, which may be costly, slow and not flexible enough. The concept of semantic trajectories has become relevant in recent years. Enriching GPS trip data with additional data helps to gain more insights into traffic behavior and can even reveal trip purposes of the drivers. This paper introduces a combination of a data-driven city segmentation with semantic trajectories. We show that enriching GPS trip data with additional information and analyzing the destination areas of these trajectories in detail helps to understand the journey and to reveal possible trip purposes. Multiple data sources are used, such as Points of Interest (POI), OD-points and whole trajectories of cars, vans and trucks. A fully automated clustering approach is introduced, which results in an interpretable city segmentation. The results are added to the trajectories before they are clustered. As a result, trajectories are clustered into four groups which can be interpreted as differences in the travel purpose as well as the start and end point of the trips. The method is demonstrated for the city of Frankfurt am Main, with trajectories, that either start, end, or pass through the city. By comparing the segmented city area with land-use maps and by interpreting a random sample of the clustered GPS trajectories, the plausibility of results is discussed.