Abstract
Monitoring and anticipating crowds are essential tasks of crowd managers. These tasks are challenging, as many factors and observations must be considered simultaneously. Methods from intelligent video analysis and crowd simulation facilitate these processes and thus offer users more cognitive space for the assessment of the whole situation. This paper discusses an approach that leverages the synergy between computer vision and crowd simulation techniques. Video footage is analyzed, delivering the number of people, their speed and walking direction. These data are then used to initialize a crowd simulation. The predictive potential of crowd simulations might help anticipate critical situations in the near future. A case study using video footage from the Hamburg Port Anniversary, a large-scale public event in Germany, demonstrates that the extracted features allow for observation-consistent initialization, which improves the reliability of the crowd simulation output. The coupling of methods not only helps to monitor current activities through video footage but also to anticipate pedestrian dynamics, thus supporting public safety.