基于数据驱动的人群仿真的方法研究与实现

发布时间：2018-06-01 01:12

本文选题：人群仿真 + 路径规划　；参考：《北京交通大学》2017年硕士论文

【摘要】：人群仿真应用于学术、商业、娱乐等领域,近年来备受关注。人群仿真的目的是对人群建模,仿真人在群体中的行为。由于人群是个复杂的自组织系统,影响行人运动的因素繁多,而且应用领域广泛,实现高质量的人群仿真效果存在着诸多挑战,这也体现了本文工作的价值。本文关注人群仿真的三个层次:全局路径规划、局部碰撞避免以及人体动画实现。首先利用Delaunay三角划分对场景建模,得到场景的静态可达路径,然后运用Dijkstra算法计算行人从起点到终点的最短路径。由于在人群运动的场景中不仅仅存在静态障碍物,所以还要考虑与周围的人避免碰撞。为了达到真实又高效的碰撞避免效果,本文重点研究基于数据驱动的仿真方法。在局部碰撞避免方法中,本文利用行人运动轨迹,建立了范例数据库;在仿真过程中,虚拟人物根据自身的状态,从数据库中查找相似的范例;通过碰撞预测,从相似范例中选择不会与其他虚拟人物或障碍物发生碰撞的行为,并使虚拟人物复制该行为。然而,此方法依赖于行人运动数据,当数据量较少时易发生碰撞现象,并且在数据量过大的情况下,由于数据搜索量增大,存在仿真效率下降的问题。针对这两个问题,本文引入计算无碰撞速度的规则和碰撞检测及解除的算法,使得在数据未覆盖仿真场景的情况下依然少有碰撞发生;利用运动轨迹数据训练人工神经网络,对仿真个体的行为进行预测,能够在仿真过程中摆脱对数据的依赖,使得仿真效率不受数据量的影响。最后利用运动捕捉数据和运动数据可视化方法生成人物的肢体动作,将行人运动轨迹扩展成人群行走动画,使仿真效果更加真实完整。从仿真的效果来看,本文工作能够有效地对场景建模,进行全局路径规划,在碰撞避免仿真方面较现有工作有很大的提升,并且为运动轨迹添加了人体动画效果。另外开发了人群仿真系统原型,将全局路径规划、局部碰撞避免和人体动画实现整合在一个系统中。原型系统实现了图形用户界面,使用户或者研究人员更方便地了解该人群仿真系统,直观地看到各仿真阶段的效果,方便开展未来的研究工作。
[Abstract]:Crowd simulation is applied in academic, commercial, entertainment and other fields, and has attracted much attention in recent years. The purpose of crowd simulation is to model and simulate the behavior of people in the crowd. Because the crowd is a complex self-organizing system, there are many factors affecting pedestrian movement, and the application field is extensive, there are many challenges to realize the high quality crowd simulation effect, which also reflects the value of the work in this paper. This paper focuses on three levels of crowd simulation: global path planning, local collision avoidance and human animation implementation. Firstly, the Delaunay triangulation is used to model the scene, and the static reachable path of the scene is obtained. Then, the Dijkstra algorithm is used to calculate the shortest path from the beginning to the end of the pedestrian. Since there are not only static obstacles in the crowd movement scene, it is also necessary to avoid collision with the people around. In order to achieve real and efficient collision avoidance, this paper focuses on data-driven simulation methods. In the method of local collision avoidance, this paper establishes a case database by using pedestrian trajectory. In the process of simulation, the virtual character looks up similar examples from the database according to his own state. Select from a similar example a behavior that does not collide with another virtual character or obstacle and make the virtual character copy the behavior. However, this method relies on pedestrian motion data. When the amount of data is small, collision will occur easily, and when the data is too large, the efficiency of simulation will decrease due to the increase of data search. Aiming at these two problems, this paper introduces the rules of calculating collision-free velocity and the algorithm of collision detection and resolution, so that there are few collisions when the data does not cover the simulation scene, and the artificial neural network is trained by moving track data. The behavior of simulation individuals can be predicted to get rid of the dependence on data in the process of simulation, so that the efficiency of simulation is not affected by the amount of data. Finally, the movement capture data and motion data visualization method are used to generate the body movements of the characters, and the pedestrian movement track is expanded into a crowd walking animation, which makes the simulation effect more real and complete. From the result of simulation, this paper can effectively model the scene, plan the global path, improve the simulation of collision avoidance greatly compared with the existing work, and add the human animation effect for the motion track. In addition, a prototype of crowd simulation system is developed, which integrates global path planning, local collision avoidance and human animation into one system. The prototype system realizes the graphical user interface, which makes the user or researcher understand the simulation system of the crowd more conveniently, see the effect of each simulation stage intuitively, and carry out the research work in the future.
【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】：TP391.41;TP183

【参考文献】