Development of a pedestrian flow simulation model with intelligent understanding of CAD plans

Abstract

Computational modeling of people’s movement in the built environment is one of the most interesting research areas attracting concern of experts from various disciplines such as architects, facility managers, environmental psychologists, etc. In the past, researchers studied pedestrian behavior mainly by intensive observations and deriving simple flow capacity theories. Recently, with the advancement of computer technology, more and more complicated computational models have been developed. Despite such technological development, many building designers of public facilities, in which a large number of people will be circulating, still use simple rules of thumb. The available crowd flow models including evacuation models and computer-based tools can seldom be used for assessing the efficiency of the built environment since they depend on which point of view in pedestrian behavior is given the main focus. The major aim of this thesis is to develop a computer simulation-based model for formulating design and management strategies for pedestrian flows in large buildings with a huge number of circulating people. The model mainly consists of three components: a data collection model, a hybrid pedestrian simulation model and a geometrical information extraction model. The author believes that a combination of these three techniques can provide an integrated solution for studying people’s movement in large buildings. Development of new algorithms and techniques for these three components constitute theoretical and practical contributions of this thesis. The data collection model relates to the parameter calibration issue, the hybrid pedestrian simulation model is the core structure involving a newly developed agent-based simulation algorithm and the geometrical information extraction model developed on the basis of graph theory concerns geometry representation (pre-processing) issue. A novel automatic pedestrian detection and tracking model has been developed, which is able to automatically collect microscopic pedestrian flow data from video. By dint of the model, video data can be automatically converted into trajectory database of each individual, thus allowing further data analysis to be performed. To address this problem, two methods have been proposed: a feature-based method and a shape-based method. The results obtained by these two methods have been compared and it has been found that the shaped-based method appear to be more accurate and robust and can produce promising results. This is due to the fact that the feature-based method depends on the quantity and quality of the sample set. This data collection model can be used for parameter calibration and verification of the simulation model. A hybrid simulation model for pedestrian flows was developed. In contrast to most of the other existing models, two levels of human behaviors were integrated in the new model: high-level route choice behavior and low-level walking behavior. The route choice behavior is modeled using a discrete choice approach, which is a classic method to select an alternative from a limited choice set based on utility maximization. As for the walking behavior model, it is a pilot work to develop a new microscopic model which proposes a new set of rules and equations for people’s locomotion. This is based on observed behaviors and the representation of the space. The walking model is agent-based, where each individual is modeled as an agent who has a set of attributes and an individual behavior model. The dynamics of individuals involves agent’s interaction with the environment and other agents. The results provided by the simulation model are analyzed and the phenomena observed when running the model are discussed. Finally, we consider the pre-processing part. When applying a microscopic model, it is always the bottleneck for application that a large amount of data has to be provided. It is always a time consuming and error-prone task in many cases. To address this problem, a novel data extraction method on the basis of graph theory was developed to automatically extract meaningful spatial/ geometrical information from the CAD architectural drawings prepared by the building designers. The information captured can be integrated with the simulation model and for post-processing presentation.

Date of Award	2 Oct 2007
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Siu Ming LO (Supervisor)