Research On Particle System Based On GPU And Its Application

Simulation of natural scenery is a hot research branch in computer graphics,which is widely applied to television advertising,real-time rendering,3D animation and games, upholster design,virtual scene and so on.It plays an important role in the information society with its unique charm.The shape of most natural landscape,such as flame,spray, rivers,waterfalls,rain,snow,fog etc,is changing stochastically,and it is difficult to model and visualize these natural sceneries by using traditional methods of modeling and simulation,so the simulation of natural scenery is also a challenging task.Particle system simulations are well suited for modeling complex phenomena such as water,cloth, explosions,fire,smoke,and clouds.They are normally realized in software as part of an interactive graphics application.The computational complexity of particle graphics simulations restricts the number of particles that can be updated in software at interactive frame rates.Based on a careful analysis on the previous related work and considering the demand of virtual scene for video game,the paper presents the design and implementation of a hardware particle system for accelerating real-time particle system.Particle system is an effective method to simulate irregular or fuzzy objects. Particle-based simulations have been increasingly used for analyzing models subjected to large deformations,especially for simulations nature phenomena such as water,fire,and smoke.In particle system,objects consist of millions of irregular particles.Every particle has lifecycle and changes shape,moves constantly,exemplifying the nature of irregular object.The theory of particle system is using numerous of elements with lifecycle and attributes to simulate fuzzy object.Every particle with attributes of size,color,velocity, lifecycle experiencing the state of producing,updating,and death.During the stage,the particle updates the attributes.In these simulations,systems of simple elements interact with environment and evolve together,subject to a set of rules.The properties and evolution of the system are also determined by randomly varying initial conditions.The visual ensemble of such a group of particle exhibits a great degree of complexity and detail, closely resembling the detail and randomness seen in nature.The merit of the particle system is the ability to simulate complex object with simple particle accurately.But in particle simulations implemented in software on a general-purpose CPU,the computational complexity of particle graphics limits the number of particles to tens of thousands.The bottlenecks are the transfer of particle data from the main processor to the graphics hardware(GPU) for rendering and the parallel computational ability of CPU.The current graphics card is designed to provide more prommability in recent years. With the rapid evolution of graphics card technology,GPU-based solutions have outperformed corresponding CPU implementations.GPU can also use any sequence of instructions to carry out programming,and make the developers to achieve the implementation of any calculation.Its efficiency in memory access,combined with an increasingly flexible programmable architecture,has turned the GPU into a valuable resource for general-purpose computations.Particle system algorithms exhibit a very high degree of parallelism that is easily uncoverd and exploited,so it is possible to exploit the GPU for implementing particle engines using the ability of implement in parallel.The merits of particle simulator running entirely on the GPU are:eliminating data transfer between CPU and GPU to increasing the number of particle;and integrating illumination and collision detection to achieve good effect.The experiment proves the accurate and realism of GPU based particle system.Collision Detection is a fundamental operation used in variety of areas such as computer graphics,computer animation,visualization and more.Collision detection and collision reaction in particle system can improve reality.Main Collision Detection method includes graphics based method(hierarchical algorithms and space algorithms)and imaged based method.Graphics based method use hierarchical tree to reduce the number of mesh to be detected.The imaged based method use depth information to detect collision.There are three basic hardware accelerate approaches based on depth buffers,stencil buffers and occlusion culling.While collision reaction for particles is a fairly simple application of Newtonian physics,collision detection can be a rather complete task because of the large number of particles and complex obstacles.This paper explores to solves the problem of inter collision detection and particles-obstacle collision detection using programmable GPU. In the paper,we use the approach of subdividing the space into regular grid to reduce the possible collision pares to detect collision between particles.And use image based method to detect collision betweeing particles and obstacle.This paper does research about the particle system based on GPU,and approve an algorithm to detect inner-collision and collision between particles and obstacle.The simulation runs entirely on the GPU thus completely eliminating data transfer between CPU and GPU.Differently from previous proposals,this paper performs numerically stable simulation of particles in confined environments,including support for detection and response of simultaneous multiple particle-environment collision detection.The work and results are as follows:Firstly,the paper introduce the the theory of particle system and the method of how to build particle system.It also analyzes the attributes of the particle and posts an algorithm about how to initialize and how to update the attributes.This paper also introduces the method of programming on GPU,and presents a new GPU-based particle simulator.This section shows data storage,data transferring between CPU and GPU,update the attributes of the particle and 3D rendering.Secondly,present an algorithm to detect inner-collision and particle-environment collision.The system uses the approach of subdividing the space into regular grid.At each grid cell,the system stores a list of all particles whose centers lie in that cell.By doing this, a given particle can only collide with the particles in the same cell or particles in direct neighboring cells,reduce the pairs to be detected.For obstacle-particle collision,the system detects the mesh and movement of particles in GPU.The problem is first converted into problem of finding intersction between a collection motion trait of particle and a set of triangles,using hierarchical tree and sub-texture to reduce the collision detection.Thirdly,simulate fountain using particle system based on GPU.Dealing with the lifecyle and death in CPU,update the attributes in GPU fragment.In order to make the simulation more realistic,adding collision detection to the system.And give an example of fountain simulated by sixty thousand particles.