"The work has been organized into working packages:
Work Package number 1: Detailed Literature Review.\
The literature has been organized into three sections: Pedestrian behaviour, pedestrian trajectory data, and modelling approaches used. For the actual references, see the pdf of the final technical document B
Pedestrian Behaviour: The pedestrian behaviour has been subdivided into normal situations and evacuation scenarios
Pedestrian trajectory data: This includes both experiments and how to obtain trajectory data by processing the raw video material
Modelling approaches: This includes cellular automata, agent-based models, and physical social-force models.
Work Package 2: Data Collection and Extraction:
In this study, we collected large amount of pedestrian data using unmanned aerial vehicle (UAV) in different European countries such as Czech Republic, Croatia, and Serbia at different events such as music festival, football matches and historical event. The videos were recorded in 4K (4096x2160) resolution at 25 fps using the consumer quadcopter DJI Phantom 4 Pro Plus. The drone was made to fly between 40 – 60 m attitude directly above the crowd (Croatia and Serbia) and slightly away from the crowd for safety issues (Czech republic) and this would eliminate occlusion effect and minimize perspective distortions.
Work Package 3: Data transformation and smoothing techniques:
The actual trajectory data extraction by AI methods has been outsourced to the UAV operating company. However, since the provided trajectories relate to the pixel positions of the heads of the pedestrians and we need the physical coordinates of the feet, i.e. the position of the actual pedestrians, we developed, calibrated and validated an affiine-linear transformation do perform the mapping from head pixel coordinates to physical feet coordinates.
Finally, we have tested and optimized different low-pass filters to eliminate high-frequency data noise while retaining the actual information
Work Package 4: Model Development and Calibration:
Models are essential in order to describe the crowd movement in a realistic way. In microscopic models, the actions, decisions and behaviours of each pedestrian as well as interactions with others are considered. We select, adapt, calibrate, and validate a variant of the social-force model (""elliptical specification"")
Work Package 5: Pedestrian Simulation and Crowd Management Strategies:
Most of the human stampedes are caused by a combination of many factors such as high crowd density, local reduction of the capacity (bottleneck), panic, etc. Therefore, an in-depth understanding of the interactions between pedestrians during normal and evacuation situations is needed for devising better crowd control and management strategies. Microscopic simulation is an appropriate tool for investigating this since it allows us to treat ‘pedestrians’ individually and their interactions. We are now in the process of developing the software. The simulation model comprises of three major components: the pedestrians, the environment (in which pedestrians are situated) and the behaviour rules (to identify how pedestrians and other objects interact with each other). Each pedestrian registers the position of obstacles, current position of other pedestrians including his/her perception about their speed and the direction of movement."
