Behavior Modeling And Simulation Technology For Virtual Crowds

The behavior modeling and simulating technology for virtual crowds simulates the perception, planning, behavior and cognitive ability of human beings and aims to reflect the characteristics of crowds realistically. Much work has been done to simulate the crowd motion, which can guarantee that the agent does not collide with other agents or other obstacles. However, it does not differentiate each agent and show the characteristics of crowds, such as the flocking, social and intelligent aspects. With aforementioned disadvantages, much research on behavior modeling methods for virtual crowds has been done from two aspects. The agent-based behavior control model studies the perception, path planning and behavior modeling for virtual crowds in different crowd densities. The behavior model based on smart event embeds the behavior options into smart event and furthermore, a computation framework has been constructed for simulating the dyadic conversations.Firstly, this paper studies the perception model for virtual crowds. Since the size and the number of perceived objects as well as the perceptual range are different when the crowd density changes, the hierarchical perception model is proposed for virtual crowds, which simulates the visual, audile, tactile and memory ability of human beings. Furthermore, this model can show the differences of the perceptual outcome in different crowd densities and therefore, it can reflect the perception characteristics of human beings more realistically.Secondly, the path planning model for virtual crowds, combining the global planning with local collision avoidance algorithms, is presented based on the observations for the planning behaviors of human beings in real life. The path planning model considers the influence of crowd density, personal space, agent attributes, and agent's role on local collision avoidance behaviors. Furthermore, sidestepping is integrated as an adaptation strategy when avoiding collisions with other agents in high crowd density. Experimental results demonstrate that the path planning model for virtual crowds can show more interesting crowd behaviors close to real life to avoid the collisions.Thirdly, the hierarchical crowd behavior model is proposed to simulate diverse and different individual behaviors as well as social and similar crowd behaviors. The model focuses on constructing―individual‖agents with different characteristics and considers the influences of personality, emotion and mood on the action selection in low crowd density; while it is more concerned with social and interactional aspects of crowd behaviors and realizes flocking behaviors and emotion sympathy in high crowd density. Therefore, it simulates more realistic crowd behaviors well.In addition,―Smart Events and Primed Agents‖organization is studied and implemented to reduce the computation complexity as well as guarantee the realism of behaviors. In―Smart Events and Primed Agents‖model, smart event informs an agent of plausible actions to undertake while agent select the most appropriate action for execution based on the priming trait and meanwhile the agent's action influences the evolution of the smart event. This scheme reduces the burden of heavy reasoning and decision-making on agents, which increases the rate of crowd simulation greatly.Finally, dyadic conversation is studied based on―smart event‖concept. The situation type, the classification of the conversation archetype, the triggering and initialization of a conversation by a conversation smart event are discussed in details to show more realistic and diverse dyadic conversations for background characters. The dyadic conversation model can increase the realism of the virtual scenario and show the social aspect of crowd behaviors. Furthermore, the integration of this model not only enriches the behavior model for virtual crowds, but also makes the users experience a more realistic scenario.