DATA-DRIVEN RECONSTRUCTION OF PROCESSES FROM PEDESTRIAN TRAJECTORIES

Agent-based simulations can be helpful in understanding the complex dynamics of human behavior. Data-driven approaches for this purpose show to be promising in extracting complex features, without relying on system-specific expert knowledge. This work aims to develop a data-driven approach that enables automatic generation of agent-based pedestrian flow models, by extracting and classifying regions of interest from trajectory data. For validation purposes, synthetic data from a pedestrian movement simulation was used for the method development. We identify stay point areas from the resulting trajectories, classify the processes occurring in these areas, and reconstruct their properties. The relevant areas and types of processes were successfully extracted in four different case scenarios. However, it is necessary to test and subsequently improve these methods by using real data. Ultimately, our methods should be applied for the automatic modeling of pedestrian behavior in critical infrastructures, such as a railway station or an airport.