Research On Physically-Based 3D Smoke Simulation And Its Acceleration Techniques

Fluid is a kind of wonderful and complex natural phenomena, just like the rising smoke, floating clouds, billowing waves, burning flames and so on. The simulation of these fluid phenomena has been widely used in the video games, film effects, virtual reality and many other fields. In the traditional flash animation, the animators have to set the fluid state of every frame by handmade. It is heavy workload and has poor sense of reality. With the rapid development of computer graphics and computational fluid dynamics, physics-based fluid animation began to spread and has been widely used. It employs the inherent physical properties of fluid and builds the physical model of fluid motion, then get the physical state parameters which describe the motion of fluid by solving the model with numerical solution, and render the scene according to these parameters to produce vivid images of smoke.This paper focuses on the smoke simulation technology which is an important part of fluid animation. We summarize various methods of smoke simulation and the recent advances in the field of physically-based smoke simulation. Additionally we introduce the navier-stokes equations in details which describe the motion of fluid, and implement a realistic and real-time smoke animation through the equations. The main contents and contributions of this paper are:(1) We solve the navier-stokes equations by the finite difference method, discretize the computational domain with collocated grid, and adopt the splitting method to simplify the equation and to decrease the calculation. To make sure stability with large time step, we use the semi-Lagrangian method to compute the advection term and use the implicit iteration method to compute the poisson equation and the diffusion term. In addition, we add vorticity confinement and buoyancy to the flow to enhance the motion detail.(2) After solving the equations, we use ray casting algorithm to render the smoke density, and adopt a fast and simple method to compute the intersection of the ray and the gird. With these methods, we generate images with high realistic effects.(3) In order to speed up the simulation efficiency, we transfer the solving and rendering process from CPU to GPU. While solving the linear equation, we design an efficient Jacobi iteration according to the GPU memory model which has low access latency and improves the performance greatly. Compared with the graphics API which used to accelerate the general computation in traditional, Our CUD A implementation makes more efficient use of GPU's hardware resources.