数据驱动的人群行为仿真建模研究

发布时间：2018-05-12 10:56

本文选题：代理模型 + 人群行为建模　；参考：《北京交通大学》2017年硕士论文

【摘要】：人的群体行为是一种错综复杂的社会现象,如何在复杂场景和事件中有效的预测和控制人群的行为,已经引起了社会学家、物理学家和计算机科学家的广泛关注,同时也成为了各国政府机构的重点关注课题。近年来,计算机建模和仿真技术在人群行为的研究工作中得到了广泛应用,并由于其良好的效果得到了众多科研工作者的青睐。虚拟人群仿真技术的研究目标是在虚拟场景中尽可能逼真地再现人群的外观、动作和行为方式,体现人群的突出特点并揭示其在运动以及行为方面的规律。在各种各样的环境下,人群会有不同的表现和行为特征,利用虚拟人群仿真技术在计算机的虚拟空间中逼真地显示大规模人群的运动过程,可以为人群运动提供分析和演示的工具。本文旨在前人工作的基础上,结合统计学、并行动力学和社会科学,利用计算机科学技术,提出一种新的普适的数据驱动的层次化方法,模拟行走人群在给定的多出入口复杂环境中的运动行为,重现其运动场景和行为特征。本文提出的建模方法主要分为两层:上层主要模拟人群的宏观行为,即行人如何进入场景并选择运动目标、行人在特定环境中的感知行为以及相应的决策和寻径过程。首先根据视频数据中提取的有效的行人运动轨迹信息,利用模糊C均值聚类算法(Fuzzy c-means algorithm,FCM),得到特定场景中行人出入场景的分布规律,进而模拟行人进入场景并选择运动目标的行为过程,然后利用文章中最新提出地数据驱动的行人运动模型,模拟行人环境感知行为和决策寻径过程;底层则主要利用互惠的速度障碍模型(Reciprocal velocity obstacles,RVO)模拟行人运动过程中的碰撞避免的微观行为。文章中提出的建模方法在爱丁堡大学的公开数据集上进行了实验验证,实验结果表明,本文提出的这种数据驱动的层次化建模方法在给定复杂环境的场景重现、人群运动行为统计和分布特征提取,以及行人轨迹预测方面效果显著,具有一定的研究价值。
[Abstract]:Human group behavior is a complicated social phenomenon. How to effectively predict and control crowd behavior in complex scenes and events has attracted widespread attention of sociologists, physicists and computer scientists. At the same time, it has also become the focus of attention of various government agencies. In recent years, computer modeling and simulation technology has been widely used in the research of crowd behavior, and has been favored by many researchers because of its good results. The research goal of virtual crowd simulation technology is to reproduce the appearance, movement and behavior of the crowd as vividly as possible in the virtual scene, to reflect the outstanding characteristics of the crowd and to reveal its movement and behavior rules. In a variety of environments, the crowd will have different performance and behavior characteristics, using virtual crowd simulation technology in the virtual space of the computer to show the movement of large populations. Can provide analysis and demonstration tools for crowd movements. On the basis of previous work, this paper proposes a new universal data-driven hierarchical method combining statistics, parallel dynamics and social sciences, using computer science and technology. The movement behavior of walking crowd in a given complex environment with multiple entrances and exits is simulated, and its motion scene and behavior characteristics are reproduced. The modeling method proposed in this paper is mainly divided into two layers: the upper layer mainly simulates the macroscopic behavior of the crowd, that is, how the pedestrian enters the scene and chooses the moving target, the perceived behavior of the pedestrian in the specific environment, and the corresponding decision-making and path finding process. Firstly, based on the effective pedestrian trajectory information extracted from video data, fuzzy c-means algorithm is used to obtain the distribution of pedestrian access scene in a specific scene. Then it simulates the behavior process of pedestrian entering scene and choosing moving target, and then simulates the pedestrian environment perception behavior and decision-making path finding process by using the newly proposed ground-driven pedestrian movement model. In the bottom layer, the Reciprocal velocity obstacle model (RVO) is mainly used to simulate the microscopic behavior of collision avoidance during pedestrian movement. The proposed modeling method is experimentally validated on the open data set of the University of Edinburgh. The experimental results show that this data-driven hierarchical modeling method is reproduced in a given complex environment. The effect of crowd movement behavior statistics, distribution feature extraction and pedestrian trajectory prediction is significant, and it has some research value.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.9

【相似文献】