Simulation Of Population Behavior Based On Data Driven

The crowd is everywhere. In a virtual environment, to enhance the credibility and authenticity of the scene, need to simulate the behavior of the crowd. Based on this demand, virtual crowd simulation technology is introduced. Real crowd scenes are composed by independent human individual, but also among individuals with independent characteristics unrelated and irregular. In order to achieve the real purpose of the simulation, researchers have proposed a number of domestic and international crowd simulation models, which are based on predefined rules or mode simulation models of complex and influenced by subjective rule definition.This thesis discusses the behavior of the population based on data-driven model is extracted scene instances from real-world environment, and examples of the redefined as a multidimensional vector stored in the database. During the follow-up of the simulation, intelligent query instance based on experience to build around obstacles scene, and then look no collision similar case in the database.Crowd simulation discussed in this thesis is still to build on the examples, but made improvements in the following areas:(1) In order to reflect the input state of complete simulation process, the locus agent over time into all the current environment to create an instance. (2) Examples of clustering algorithms to extract clustering, build hierarchical database instance. Each cluster represents a class of similar patterns of behavior. (3) the use of artificial neural network trained clustering result, to obtain the corresponding relationship between instances and clustering index.We also discussed the role of A-Star global path planning and barrier methods based on the speed of local path planning model in addressing the crowd behavior simulation and collision avoidance behavior personalized consideration intelligence body to a local collision avoidance process in order to achieve real reflect the purpose of the crowd.Finally, through the experimental comparison of our model and RV02 crowd simulation model, they have a high accuracy trajectory elongated corridor at the scene, they have a lower trajectory information mean and variance; in the simulation efficiency,2-5 agents through the elongated hallway scene has 13%of the average time promotion. We also simulating agent in individualized behavior sideways, agent peers by experiments, it can be drawn:collision avoidance measures in pedestrian behavior under the influence of personality can be more truly reflect crowd behavior.