doc. Ing. Pavel Hrabák, Ph.D.

Specialist supervisor: Ing. Hana Najmanová, Ph.D., prof. RNDr. Pavel Surynek, Ph.D.

Multi-agent (MA) models of pedestrian dynamics become a promising tool supporting an evacuation analysis in performance-based fire safety design. A pedestrian or evacuee is represented in such model by an agent, which parameters, dynamical and interaction rules are inspired by the physical, cognitive, and psychical processes of people in crowd. Recent studies show that very important but rarely considered aspect influencing the evacuation process is the heterogeneity of the crowd in various levels. Most challenging, and so far poorly discussed, is a simulation of a crowd composed of evacuees with essentially different roles and competences during the evacuation (following or giving instructions, keeping or controlling the formation, etc.) as e.g. evacuation of school (pupils vs. teachers) or cultural event (visitors vs. staff).

Goal of the thesis is to fill in this research gap in several of the following aspects:

• Thorough investigation of the influence of essential agents’ heterogeneity on important characteristics of pedestrian flow and evacuation process in most common MA models (cellular, social-force, or rule based).
• Development of new or modification of existing MA models enabling the above-mentioned hierarchical heterogeneity, supporting inter-agent bonds and crowd formations.
• Organization and analysis of evacuation experiments addressing the above-mentioned issues. Comparison of experiments and simulations, development of metrics quantifying the correspondence of heterogeneity in models and reality.
• Investigation of possible strategies of leading/controlling pedestrians/agents influencing the evacuation process by means of the multi-agent path finding or planning tools.

The work should be consulted with both, fire engineering expert and expert on multi-agent planning processes.

doc. Ing. Pavel Hrabák, Ph.D.

Ing. Daniel Vašata, Ph.D.

Department of Applied Mathematics

The fundamental diagram is a crucial characteristic of pedestrian flow that enables engineers to judge the maximum capacity of pedestrian facilities. The classical methods of estimating the fundamental diagram use analytical approximations of this relationship. However, they fail to take into consideration the specific properties of each geometry and the composition of the pedestrians. This thesis presents a data-driven approach to estimating the fundamental diagram via artificial neural networks. This includes simulating evacuations in an open-source pedestrian simulator, computing an alternative measurement of density - the mean distance to k-nearest neighbors, and subsequently training a neural network to estimate the fundamental relationship between density and velocity. The trained models show significant ability to extract the macroscopic relationship from the individual measurements. The prediction is then expanded to include more explanatory variables - the relative positions of nearest neighbors and the relative distance to the exit. This further reduces the prediction error, demonstrating the applicability of machine-learning methods in the field of pedestrian dynamics.

Thesis on DSpace

doc. Ing. Pavel Hrabák, Ph.D.

prof. RNDr. Pavel Surynek, Ph.D.

Department of Applied Mathematics

Agent-based cellular models can be used to simulate the process of evacuation of people from a room. The actions and interactions of heterogeneous agents create collective motion and capture complex phenomena of pedestrian dynamics. This thesis presents a multi-agent cellular model based on floor-field model and is extended by a new strategy for solving conflicts when two or more agents attempt to enter the same cell. The agents and the model have various parameters that influence the conflict solution. A sensitivity analysis on these parameters is performed that reveals the individual contribution of variance in the results.

Thesis on DSpace

doc. Ing. Pavel Hrabák, Ph.D.

Ing. Jakub Novák

Department of Applied Mathematics

The thesis focuses on processing and analysis of a video record from crossing pedestrian flows experiment conducted at CTU FNSPE in 2014. The result is an algorithm for automatic extraction of trajectories from this video. Pedestrians in the video had special hats for recognition. The tracking of people is based on hats detection from video frames. Track identity association is done using the shortest distance. When tracking, it can happen that part of the trajectory is missing. The missing parts are approximated by a line segment. Next aim is to recognize binary code from hats. With usage of a convolutional neural network 45\% accuracy was achieved on 20 randomly picked hat samples. The outcome of the thesis is a dataset of trajectories and its analysis using pedestrian flow characteristics (average speed, velocity, density, and fundamental diagram).

Thesis on DSpace

doc. Ing. Pavel Hrabák, Ph.D.

Ing. Jakub Novák

Department of Applied Mathematics

This bachelor thesis focuses on the processing of data obtained by the experiment of simulated evacuation of people through narrowed section. Multiple video recordings were made of the experiment, frow which is needed to extract the trajectories of individual's movements using neural networks and computer vision, and then compare and analyze them. The result of the thesis is to obtain trajectories of pedestrians, their comparison with the trajectories obtained by organisator of the experiment, Ing. Marek Bukáček, and to perform statistical analysis of trajectories.

Thesis on DSpace

doc. Ing. Pavel Hrabák, Ph.D.

Ing. Daniel Vašata, Ph.D.

Department of Applied Mathematics

Time headway in bottlenecks is an important characteristic of pedestrian behaviour. This thesis reproduces the result of previous research in this field, where several statistical models for prediction of time headway distribution in bottleneck based on walker's surroundings were proposed. I estimate parameters of these models on data from evacuation experiments organized by CTU and discuss the differences with results from previous research.

Thesis on DSpace

doc. Ing. Pavel Hrabák, Ph.D.

Ing. Jana Vacková, Ph.D.

Department of Applied Mathematics

This bachelor thesis presents a stochastic agent-based model that simulates pedestrian movement in the Luzen Valley, National Park Šumava. The model merges three distinct mathematical models. Random arrivals of agents are modeled by means of a Poisson process. Further, a fundamental diagram-based model is introduced to transport agents to the estimated bottleneck of the trail. The third microscopical model represents a movement of pedestrians on the bottleneck. The transport section introduces the agent mass in the perceived surroundings of the agent as a novel concept to model interaction between agents that represent pedestrians. Furthermore, kernel estimates are implemented as an individual mass generated by each agent. Due to the focus of this thesis on the transport section of the model, the bottleneck section is represented by a very simple cellular automaton, a totally asymmetric simple exclusion process. Further, this thesis successfully implemented the model in a simulation tool that has been developed to conduct simulation experiments with the model. The model parameters were calibrated using empirical findings from pedestrian movement. Correctness of kernel implementation was confirmed by one of the experiments. Furthermore, the experiments with the arrival intensity revealed that jams occur at the average arrival intensity of 48 pedestrians per minute. This value corroborates the reasonableness of the transport section model, as it aligns with the empirical findings that were not used for calibration. Additionally, the experiment results validated the assumption that heterogeneity in speed of agents and heterogeneity in arrival intensity leads to earlier stoppages.

Thesis on DSpace

doc. Ing. Pavel Hrabák, Ph.D.

Ing. Daniel Vašata, Ph.D.

Department of Applied Mathematics

The influence of exit definition on the pedestrian behavior during evacuation is studied via a simulation toolkit based on cellular model. This thesis presents modified Floor-Field model that supports different types of exits. Simulation results compare the influence of exits on different agent configurations: homogeneous, when all agents are equivalent, and heterogeneous, when agents are divided into different groups. The comparison shows that the heterogeneity in aggressiveness and sensitivity to occupation can change the properties of the flow, causing it to lose stationarity and turning it into a linearly decreasing flow, and changing the type of exit in this case can enhance this effect.

Thesis on DSpace

doc. Ing. Pavel Hrabák, Ph.D.

Ing. Daniel Vašata, Ph.D.

Department of Applied Mathematics

The thesis' main goal is to identify the most influential variables in determining the total evacuation time of a simulated egress scenario. Two contrasting geometries are investigated - a double-decker train and a lecture hall. For each of these geometries, two distinct scenarios are considered - a low-density setting where the facility is at approximately 50% capacity, and a high-density setting where the facility is at 90% capacity. These scenarios are analyzed by means of sensitivity analysis with a special focus on the individual sources of randomness in the output - the occupant parameters, their initial positions, and the inherent uncertainty of the simulation process. The crowds in the experiments are made up of two groups of occupants - an able-bodied group with no movement limitations, and a movement-impaired group simulated by a slower maximum speed and a higher acceleration time. The results show that at lower densities, the initial positions of the occupants play a crucial role in determining the total evacuation time, but at higher densities, the macroscopic characteristics of the crowd take over as the most important factors.

Thesis on DSpace

doc. Ing. Pavel Hrabák, Ph.D.

prof. RNDr. Pavel Surynek, Ph.D.

Department of Applied Mathematics

Children, who account for a significant part of the population, are among the most vulnerable during evacuation, along with the sick, elderly, and disabled. Recent experimental research conducted in preschools in the Czech Republic has provided a theoretical foundation for developing a simulation model that captures children's specific behavior during evacuation. The model proposed in this thesis is a hierarchical system of leading and following agents coordinated by a central planning algorithm. It is based on an extended floor field model. The main contribution of this work lies in analyzing novel strategies and rules for the leading agent, such as it's position within the crowd. The crowd structure is modeled by introducing a pair formation of children moving in queues and examining the leader's influence over the following agents. Finally, the thesis discusses limitations and suggests possible areas for future improvement.

Thesis on DSpace