群体行为规划技术的研究与实现

【作者】 罗伟平；

【导师】 谭浩；

【作者基本信息】 电子科技大学 ， 计算机应用技术， 2010， 硕士

【摘要】 群体动画在计算机游戏、影视动漫、城市建筑规划等很多方面都有着广泛的应用。然而,对大规模群体行为进行模拟尤其是实时模拟是一件复杂而艰巨的工作。因为对群体行为的仿真不仅涉及到高层的决策过程,还同时要考虑底层事物的表示以及事物之间的交互计算。更为麻烦的是,个体与个体之间以及个体与环境之间的复杂约束关系决定了在群体规模增大的时候,仿真运算复杂度将呈非线性增长。为此,本文提出构建一个高效的群体行为引擎来解决这一问题。本文首先讨论了行为系统的设计,然后着重研究了在路径规划算法方面进行改进及系统实现中使用加速技术来提高系统性能,最后介绍了群体的建模方法以及行为引擎如何被用来进行快速实现。构建行为引擎的目的在于简化群体仿真应用的开发过程。它的难点在于如何对行为仿真中的各种元素进行抽象、组织和融合。本文将行为引擎设计为信息管理模块、路径规划模块,以及系统控制模块三分部分。每个功能模块的作用和设计都做了比较详细的讲解。它预置了一些关键算法来提高行为仿真质量,同时也给用户提供了很好的扩展接口。路径规划是大规模群体仿真性能消耗的最主要部分。本文提出了一种基于势能场的实时高效的群体导航方法。该方法将局部势能场与全局势能场相结合,有效的解决了局部势能场中的极小值问题,同时又避免了全局势能场无法满足局部灵活性的问题。并行计算技术是目前最为流行的计算加速方法之一。本文分析了如何对碰撞避免和行为个体的更新进行适当修改,以适应并行执行。同时介绍了OpenMP并行方案如何被用在行为引擎中实现并行运算。另外,本文探讨了另一种简化运算的方法——LOD技术,并介绍了它如何用在行为仿真中以简化运算。最后,本文介绍了群体行为的建模方法,并以交通仿真和室外逃生系统为例,将重点放在如何利用行为引擎系统进行快速的群体建模实现,同时也作为行为引擎系统有效性的验证。随后,本文对目前的研究成果进行了总结,并对将来进一步的工作进行了展望。更多还原

【Abstract】 In computer-games, films and urban-designing, crowd simulation with computers have a wide range of applications. However, large-scale crowd simulation of behavior, especially real-time simulation, is a complex and arduous task. The crowd simulation of behavior involves not only high-level decision-making process, but also taking into account that the bottom of things, as well as the calculation of the interaction between objects. The more troublesome is that the complex relationships of objects make the simulation show a non-linear increase in computational complexity with crowd size get larger. This paper proposes to build an efficient behavior engine to solve this problem. First, this paper discusses the designing of behavior system. And then this paper focused on studying aspects of the path planning algorithm and system implementation with acceleration techniques to improve system performance. Finally, this paper introduced the methods of crowd modeling and how the behavior engine to be used for fast implementing it.The behavior engine aims to simplify the development process of application about crowd simulation. Its difficulty lies in how to abstract and organize the various elements of simulation. This paper will divide the engine into three parts: the module of information management, the module of path planning and the module of controlling. Each functional module has a detailed explanation. The engine preloads a number of key algorithms to improve the quality of simulation, and also provides good extensions for users.Path planning is the most consumed part of the simulation performance. This paper proposes a real-time and efficient navigation method with potential. The method combines the local potential with the global potential. It is an effective solution of the minimum value problem with local potential, while avoiding global potential can not meet the local flexibility.Parallel computing technologies are currently the most popular one of the methods of calculation speedup. This paper analyzes how to use parallel computing technologies for the collision avoidance and behavior of individual updates. The paper also introduced how to use OpenMP technology to develop the engine for realization of parallel computing. In addition, the paper explores another method of simplified operation - LOD techniques, and describes how it is used in the behavior simulation to simplify the operation.Finally, the paper describes the method of modeling crowd behavior to implement traffic simulation and outdoor escape system. It focuses on how to implement rapidly crowd model with the engine system. It is also as a validation of the effectiveness of engine systems. Subsequently, this paper summarizes the research results and looks to further work.更多还原