Dynamic data - driven simulation of pedestrian movement with automatic validation Jakub Porzycki 1 , Robert Luba´ s 1 , Marcin Mycek 1 , and Jaroslaw W ˛ as 1 Abstract The article presents a dynamic data-driven simulation of pedestrian move- ment based on the generalized Social Distances Model, where simulation system is continuously synchronized with current flow data, gained from Microsoft Kinect depth map. Both simulation and data analysis are real-time processes. Agent appears in simulation, as soon as consecutive pedestrians leave sensors tracking zone. Due to system architecture containing feedback loop, automatic validation and parame- ters calibration is possible. A new method of depth map based pedestrian tracking is proposed as well as a new algorithm of pedestrian parameters extraction for short trajectories. Paper describes in details proposed algorithms, system architecture and an illustrative experiment. 1 Introduction Using of the paradigm of data-driven simulation means, that the created simulations are influenced online by the real data, not only by offline parameters or personal ex- perience and intuition of the authors. Thus, in data-driven approach, the simulation’s input should be continually fed with actual data [5]. In crowd dynamics simulations, the sources of the data are (most commonly) attributes of pedestrians extracted from the video recordings, mobile phones or, most recently, from other electronic devices like Microsoft Kinect. Thanks to such approach, it is possible to prepare short-term predictions of crowd behavior in specific, well-defined situations. Jakub Porzycki, Marcin Mycek, Robert Luba´ s, Jaroslaw W ˛ as Department of Applied Computer Science AGH University of Science and Technology, al. Mickiewicza 30, 30-059 Kraków, Poland, e-mail: {porzycki, rlubas, mycek, jarek}@agh.edu.pl 1