The Research On Real-time Simulation Of Massive Crowd Motion

With the rise of virtual reality and the development of crowd motion simulationtechnology, massive crowd motion simulation has started to become the new focus. Ithas been a critical topic to study and solve the problems more completely, accuratelyand carefully, which makes crowd simulation have a higher efficiency and betterrendering. Currently, there have been a lot of achievements, but some problems stillexist to construct a general crowd simulation model. Deeper explorations are neededfor this research, so the research on it has both scientific significance and applicationvalues.In recent years, crowd motion simulation technology has been widely used inmany fields, including simulation training, film special effects and public safe-aideddesign, etc. In addition, people place greater requirements on the sense of reality andsimulation efficiency with the development of computer technolo gy. Therefore, inorder to simulate crowd motion efficiently and veritably, we study the massive crowdand put forward a general crowd simulation framework aiming at reaching thedemand of both efficiency and reality. The specific work done is as following:(1) We propose a crowd motion planning approach based on discrete field. Itconstructs guidance map saving the information of the environment to implementglobal path planning, and then constructs navigation field to finish local path planningusing guidance map. After generating the crowd motion path, we make the pathsmoother for higher realism. Additionally, density field is introduced to control thespeed of the crowd to prepare avoiding collision between individuals.(2) An interactive behavior model based on a user-defined interactive rule is alsopresented and it is oriented to group interaction and individual interaction. Groupinteraction is controlled by the behavior primitives and individual interaction iscontrolled by individual’s motion states and the interactive behavior selectionprinciple. Moreover, the collision avoidance between individuals is implemented bythe method based on the dynamic space filling according to the density of the crowd.(3) We design a crowd motion simulation system which is implemented byseveral different architectures taking advantage of the features of CPU and GPU.Through a great number of experiments, we compare and analyze their performance.Then, we achieve the crowd simulation system based on heterogeneous parallelarchitecture with multi-cores of CPU and GPU and it can get the best performance.Finally, the crowd simulation system by applying the proposed approach andarchitecture obtains good rendering effect and real-time simulation efficiency.