Zhang, ZilinCrowds with densities higher than 2 persons/m2 can be defined as high-density crowds. High-density crowds presenting in daily life like concerts and sports events can lead to serious people and property loss. This research work focuses on developing a new speed-density relationship for high-density crowds. The applicability of the new speed-density relationship is tested in Pathfinder, an agent-based evacuation simulation software developed by Thunderhead Engineering, to determine the impact of updating such data on the model's performance. This research also discusses several parameters and functions in Pathfinder including acceleration time and reduction factor to help model high-density crowds.Previous work is available for analyzing crowds with densities lower than 3 persons/m2. However, densities as high as 9 persons/m2 are common in many high-density crowd scenarios. The disparity between previous work and real-world situations presents a challenge for engineers to understand the crowd dynamics of high-density crowds. Developing evacuation models to predict the behavior of high-density crowds is crucial to improving the predictive ability of crowd simulation. By doing so, it helps to reduce the number of casualties in future emergencies. Real-world footage videos are analyzed in this research. With the open-source experimental footage videos provided by Jülich, a national research institution, a new speed-density curve is summarized by collecting and analyzing data from the videos. The assessment of the applicability of the new speed-density in Pathfinder focuses on four aspects: evacuation time, flow density, flow velocity, and occupants’ arrangement. Per case examined, by applying the new speed-density curve, the predicted evacuation time from Pathfinder simulation is improved from 12.6% to within 4.9% of the experimental video time. The predicted flow density is improved from 6.2% to within 0.7% of the average video density. The predicted flow velocity is improved from 25.9% to within 3.3% of the average video velocity. At the same time, it is observed that occupants in the model behave more realistically.enThesisEngineering