Researches On Key Technologies And Military Application Of Crowd Behavior Simulation

With the increase of social unrest and frequent occurrence of emergency security incidents, the domestic group events have caused tremendous damage and casualties to society. In this context, improving the ability of responding to emergencies is important to ensuring public safety. In recent years, using computer simulation technology to the studies of crowd behavior simulation and providing technical supports for crowd management control decisions with a military application background have become research hotspots.This thesis takes group events as the representative background of military application. Currently generating realistic crowd behavior model is encumbered by the difficulties such as current model has poor authenticity, a single model can’t simulate the phenomenon of crowd behavior realistically, the simulation framework possesses bad adaptability, as well as there is some actual demands of group events on crowd behavior models. We conduct in-depth researches on the key technology of the virtual crowd behavior simulation from the following aspects, such as architecture design, independent individuals’ navigation behavior, generating various types of group behavior, control behavior in the group events group and so on. The corresponding control strategy and model are proposed. And the traditional algorithms are improved. The main contents of the dissertation are summarized as follows:(1) Firstly, the virtual crowd behavior modeling technologies and applications are summarized, the advantages and disadvantages of different methods are analyzed and compared. The group events are regarded as the application background. The crowd is divided into civilians and soldier, and the civilians consist of individual and group. Thus By modeling the different constituents of the crowd, a realistic crowd behavior model is established.(2) For the problem of designing the crowd behavioral simulation framework, a multi-agent architecture is employed. A behavioral/cognitive hybrid architecture is presented by combining the behavior-based agent architecture and deliberative architecture. This architecture not only can design complex individual behaviors which are significantly different from the rules of reaction, but also generate more complex behaviors by adding a certain degree of planning module to the behavior modules. So it is well adapted to unstructured environment by achieving complementary advantages of complex high-level planning and rapid responses to environmental changes. Established on behavior-based system framework, behavioral/cognitive hybrid control structure has more advantages than deliberative/reactive hybrid architecture or the one which is constructed only on the behavior-based structure. Consequently, it is more suitable for crowd behavior simulation. At the same time, in our architecture, three combinations are achieved:one is the combination of global planning and local behaviors, another is the combination of macroscopic and microscopic behaviors, and the other is the combination of data-driven approaches and model-driven approaches. Our behavioral/cognitive hybrid architecture greatly enhanced the adaptability and expansibility of crowd behavior simulation framework.(3) For the problem of generating independent individual’s navigation behavior to improve the realism and naturalness of virtual human’s trajectories, the trajectories must satisfy smoothness, diversity, and has maximum speed and acceleration constraints. Firstly sampling strategies of PRM (probabilistic roadmap) are modified adaptively. Non-uniform sampling and bridge-test sampling strategy are combined and selected as the sampling strategy of the virtual human. Based on decoupled trajectory planning method, a diversity trajectory design strategy which has the speed and acceleration constraint is proposed. Meanwhile, in order to improve the authenticity of the avoidance behavior, the virtual human’s emotions in psychology, roles in sociology and personality parameters are integrated into the avoidance behaviors. Then diverse collision avoidances are generated depending on these intrinsic properties of virtual humans.(4) In order to generate realistic group behaviors, the Visibility-Voronoi diagram is used to generate the motion paths of the groups. Then smoothing methods, diversity trajectory generation strategy and a modified evaluation function are utilized to generate the group’s trajectories. So that the group’s trajectories can meet the needs of four aspects, i.e., there is a specified clearance between the group and the obstacle, the group’s path should be as short as possible, the group’s path should be smooth and possess diversities.For non-coherent group, the basic flocking behaviors are analyzed based on distributed control algorithm. Then a flocking method with multiple leaders and global trajectory constraints is proposed by designing a two-stage control strategy. In the first stage, an imaginary point which serves as a virtual leader is set on the desired trajectory. It is responsible for ensuring that the center and average velocity of all the leaders in the group converge to the global trajectory. The second stage is responsible for making the center of the group satisfy constraints of the global trajectory by using the leader agents to lead the follower ones. And ultimately the center of the group meets the global trajectory constraints by combining these two stages. In addition, by introducing a class of "imaginary β agent" on the border of the barriers, the group possesses collision avoidance capability from any obstacle.Aiming at the application for group events, the behaviors of the coherent group of civilians and soldiers are studied. The contour of the group is used to keep the group’s coherence. And it can adapt to the environment and curvature of the global trajectory flexibly. And the behaviors of a group of soldiers which have strict formatior constraints are modeled by integrating the group’s contour and formation. In this process, a formation-key-frame sequence is constructed on the basis of the group’s contour and the desired target formation so that all the soldiers in the group move along the global trajectory from the initial formation in the initial area to the target formation in the target area. On one hand it can ensure that there are no collisions between individuals, on the other hand it can ensure that the formation of the group approaches the desired formation as much as possible and the formation has adequate flexibility.Altogether, the proposed method can produce a variety of group behaviors in crowd behavior simulation to meet the needs on it in daily lives or group events.(5) In order to achieve crowd managements in daily life and riot controls in group events, an improved shepherding behavior is proposed for group control problem. According to the deficiency of existing shepherding behaviors, the time optimal trajectory planning algorithm with maximum speed and maximum acceleration constraints is introduced to generate the trajectories of the control agents. Consequently, on the premise of meeting the maximum speed and maximum acceleration constraints, the control agents move to target position as quickly as possible, and thus manage and control the populations more effectively.(6) Finally, the virtual crowd behavioral simulation system and its application are developed based on the hybrid data- and model-driven methods. On the basis of the currently available virtual character software, the velocity field based method and the agent-based method are integrated to achieve both complementary advantages. The widely used business software DI_Guy and Vega Prime software platform are utilized as a visual rendering output platform. And the bottom SDK action library functions are called to control the characters in virtual crowd. Then taking the scenarios of group events as examples, the relevant application systems are fulfilled. Then the research works done in this paper are tested. Simulation results show that the proposed method can simulate various crowd behaviors in group events realistically. Thus, it can provide valuable references for crowd managements and crowd behavior analyses in group events.