Spatio-temporal patterns and pedestrian simulation

• Main Author: Hu, Nan
• Other Authors: Cai Wentong
• Format: Theses and Dissertations
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; DRNTU::Engineering::Computer science and engineering::Computing methodologies::Simulation and modeling
• Online Access: https://hdl.handle.net/10356/61133
• Institution: Nanyang Technological University

Pedestrian behavior is an interesting social phenomenon. Understanding such crowd behavior has many important applications in computer animation, digital entertainment, civil planning, military training, etc. Despite significant research and development efforts, it still remains a challenging task. Thorough understanding of pedestrian behaviors requires true multiple disciplinary scientific research. With the fast development of computational technologies, modeling and simulation has become an important approach to understanding crowd dynamics. This thesis continues this trend and aims to model human-like steering behaviors of pedestrians. While simulation is an attractive approach to understanding crowd dynamics, evaluation, in particular, validation of pedestrian models is very much an open research question. This the- sis also contributes to this area by the development and application of a three-stage validation method. The thesis starts with a comprehensive review of both pedestrian studies from social and psychological research, and existing modeling and simulation techniques. This work provides a theoretical basis to understand pedestrians’ navigational behaviors and thus helps support a computational model for realistic simulation. Based on existing pedestrian studies and observations on pedestrians’ steering behaviors in real life situations, a tri－layered modeling structure of their navigational behavior is proposed, which focuses on modeling the middle layer between the traditional path planning and locomotion layers. This middle layer characterizes some commonly ob- served steering behaviors as steering strategies. A pattern-based approach is used to mimic the cognitive process of pedestrians. This approach describes: a perception process which pedestrians use to asses the situation through Spatio-Temporal Pattern (STP); a method for choosing corresponding steering strategies; and finally a way to execute these strategies. A small number of commonly observed steering strategies are used to produce complex steering behaviors. Decisions on the selection, scheduling and execution of steering strategies in a given situation are based on the matching result between the currently perceived concrete STP and the prototypical STPs stored in an agent’s experience base. The STP representation and a hierarchical matching mechanism have also been proposed to formulate the experience of a pedestrian and to imitate the cognitive process involved in their steering process. Validation of a computational model is a challenging task in pedestrian simulation, and it is still an emerging research topic in this area. To bridge this gap and to evaluate the proposed model in a systematic manner, a three-stage validation method is proposed through: (1) comparison studies against controlled field experiments, (2) model comparisons in standard test cases adopted from the literature, and (3) a sensitivity analysis of the model parameters. Such a validation methodology is based on a comprehensive review of evaluation techniques used in crowd simulation. Existing validation types and techniques are first categorized along different dimensions. A sequence of validation processes for different validation types is then conducted in three stages. The results are analyzed and demonstrate that the proposed model (1) is capable of generating quantitatively more similar simulation results to the experimental data than other comparing models in selected scenarios; (2) presents various crowd phenomena as reported in the literature, reproduces the collective efficiency and various crowd dynamics under different complex scenarios and (3) is robust to generate simulation results according to different key parameter values. Such results suggest that the proposed approach fulfills the objectives of generating realistic steer- ing behavior of pedestrians in various situations. In summary, the contributions of this thesis primarily fall into three parts. First, a tri-layer motion modeling structure to simulate the navigational behaviors of pedestrians has been proposed. In addition to the traditional two layered motion modeling system through path planning and locomotion control, a strategic steering behavioral layer in the middle of the traditional two has been explored in the thesis. Secondly, a novel pattern-based approach is proposed to model pedestrians steering behaviors realistically at this layer. A pattern-description method is proposed to describe and specify the novel spatio-temporal patterns, and a hierarchical pattern-matching process is developed to mimic the decision-making process. Thirdly, a comprehensive review of the model validation methods for crowd simulation has been discussed. The result is a novel three-stage validation methodology, which in turn is applied to the proposed model and demonstrates the model is capable of producing realistic crowd dynamics.