Research On Behavior Modeling And Control Methods For The Crowd In Mass Disturbance

In recent years,mass disturbances caused by a variety of social threat factors have occurred frequently,which are severe threats to the personal and property security of a large number of citizens.In consequence,the group behavior modeling has drawn more and more attention in the field of crowd safety and management.However,the researches on the mass disturbances in complex urban environment are still insufficient,and more real and effective crowd modeling and simulation methods are needed to provide guidance and supports for the government's security department in the crowd management and crowd evacuation.So far,the application range of simulation frameworks for the crowd behavior simulation is still narrow,and the diversity and authenticity of the behavior models are not good enough and the control methods for crowd behaviors need to be improved.Based on the above background,from the aspects of the actual needs of crowd simulation in mass disturbances and the virtual human's sociality and intelligent,the in-depth researches on the behavior modeling and control strategies for the virtual crowd are conducted in this dissertation,which are mainly simulation framework design,agent navigation behavior,group behavior generation and control respectively,then the corresponding behavior models and improved algorithms are proposed.Concretely speaking,the main contents of the dissertation are the following four aspects:(1)Firstly,the research background and significance of the behavior modeling and control methods for virtual crowd are outlined.Then,we adopt the behavior-based architecture to develop the crowd behavior simulation framework for mass disturbances,and five basic components of the framework are designed in detail,which are Environment Generator,Population Generator,Attributes Renovator,Crowd Behavior Generator,and Visualization Engine.In the Crowd Behavior Generator,the Basis Behavior Repertories are implemented correspond to different group types.Thus,the crowd phenomena generated by the crowd behavior simulation framework is more realistic.This simulation framework can be used to support the behavior simulation for the stabilisation groups,civilian groups,and rioter groups in the urban environment,which provides the theoretical foundation for the behavior models and control algorithms in crowd simulation.(2)Based on the Probabilistic Roadmap Method(PRM)and collision avoidance behavior,the navigation behavior and motion planning of virtual human individuals in dynamic environments are studied.To generate the navigation behavior of the urgent members under the dynamic constraints,an optimal acceleration-velocity-bounded trajectory planning method based on a given path is presented for environments with dynamic obstacles.This trajectory planner ensures that the moving time of the urgent members is minimal by using a velocity interval propagation algorithm to compute the reachable velocity sets at obstacle vertices in state-time space.Finally,combining with the collision avoidance behavior and the flocking behavior,the crowd simulation with dynamic obstacles are implemented.The simulation results demonstrate that the navigation behavior is effective and the generated crowd phenomenon is realistic.(3)For the group behavior generation and motion planning,the groups of the crowd are divided into coherent groups and non-coherent groups,and the behavior generation methods for each type are studied respectively.For the non-coherent groups,the separation properties of the groups with more than 4 members are analyzed,and it is assumed that the group will be separated into multiple small groups with 2 to 4 members during the movement.Two rules are designed to simulate the consistency and adhesion of these separated sub-groups.For the small groups with 2 to 4 members,a group formation generation and motion planning method is presented.In this method,a set of basic group formations is established,and then a series of candidate formations can be generated by interpolating current group formation with the basic formations.After that,a collision avoidance behavior is used to compute the candidate velocity set for each candidate formation.Finally,an improved evaluation function is used to choose the appropriate group formation and group velocity from the candidates at the next simulation step.Then,the actual velocity of each group member can be calculated.The group members generated by our method can adjust their formations automatically according to the changing environment and accomplish collision-free movement to reach the goal region.For coherent groups with constant area,a novel behavior generation method is proposed,and it integrates the C-L method into the probabilistic roadmap algorithm with sampling on the medial axis.In the preprocessing phase,the group is discretized into a grid-set which represents the configuration of the group.Then,a number of samples are generated on workspace by the medial axis technique.These samples are extended into group's configuration nodes of the roadmap using an extending strategy.Meanwhile,the group's deformation degree relative to the desired shape is introduced to improve the evaluation function.The evaluation function gives users more flexibility to determine the respective weights between the group's deformation degree and its distance to the goal in the query phase.After that,a novel local planner using C-L method and the improved evaluation function is constructed to connect any two neighbor configurations.Experiments show that the proposed approach is able to generate reasonable behaviors for the coherent group efficiently and maintain its consistency when moving toward the goal.(4)Based on the analysis of the behavior generation methods for different types of groups,we design and improve the Probabilistic Roadmap Method and Shepherding behavior to study the group behavior control methods.To simulate the control abilities from stabilisation groups to the civilian groups and rioter groups in mass disturbances,an improved Shepherding behavior is proposed by introducing a module,which is used to decide when to control the flock according to its state.And the approaching behavior of the shepherds is implemented by using the acceleration-velocity-bounded individual's navigation behavior,which can lead the shepherds to arrive at the steering points in shortest time under the dynamic constraints.Moreover,the steering behavior of the multiple shepherds is also improved.Thus,the improved Shepherding behavior can be used to simulate the control behaviors in mass disturbances.For the Protection behavior,the separation definition of the flock and the allocation rules of the shepherds are analyzed,and then the improved Shepherding behavior is utilized to achieve the behavior that the stabilisation groups can escort the scattered civilian groups to a designated safe area.Finally,the simulation results show that the group behavior control method is effective.In conclusion,the work in this dissertation can be used to analyze and simulate the crowd behaviors in certain environments for the maintenance of social stability,and it can be utilized as supplementary practices to improve the management and control ability of the stabilization officers to various groups in mass disturbances.