The Real-time Virtual Fish System Based On GPU-acceleration

Real-time simulation for movement of fish on computer is always a hot topic in computergraphics, especially in games, the special effects of film and electronical products forentertainment. Not only realistic appearance, but also physical and real-time movement, aredemanded by people. However, the practical movement of fish is a complex process, which isoften hard to descript through some simple processes. How to find these laws that cansimulate movement of fish on computer becomes a key problem. Nevertheless, on one hand,there are no appropriate models or description for movement of fish in current research areafor graphics, on the other hand, the complexity of the physical computation on movement offish often leads to serious performance degradation which cannot be adapted to thoseapplication areas sensitive to time. So it is specially important to study the accelerationtechniques within the human's visual error. Due to the above reasons, the objective of thispaper is to simulate movement of fish realistically and fluently on common PC. UsingGraphics Processing Unit(GPU) as a tool, this paper concentrates on finding the tradeoffbetween the realistic effects and real time performance, to meet the different needs fromvarious applications.In computer simulation animation system, simulation for movement, collision detectionand texture process are three key techniques. This paper studies these issues by virtual fishsystem and lays a strong emphasis on the following three points: 1. We study physics basedcomputer animation technology, design simulation methods fish of fish movement based onuniversal movement model and GPU acceleration axial deformation, and realize real-timesimulation for fish movement on common PC; 2. We study collision detection techniques fordeformable objects, and propose an efficient algorithm for collision detection based onparticle swarm optimization(PSO) and GPU, which improves the frame rate. 3. We studytexture mapping and color transfer in 3D modeling, present a color transfer method that canbe conveniently used for generating kinds of fish skin texture.The main contribution of this paper can be generalized as follows:(1) For solving those problems in conventional simulation for movement of fish, such asthe poor universality, the large computation on grid models, difficult realization of real-timesimulation on common PC, this paper proposes a new axis deformation algorithm based GPUfor real-time simulation of fish movement. Firstly, on the basis of the analysis of the biologymechanics, we divide the fish body into several sub-modules, and parameterize the fish movefunction. Secondly, we present the grid model axis deformation algorithm based on GPU,which is a combination with move function in fish simulation. Compared with other currentsimulation methods for fish movement, this method has better expansibility, and simulates easily movements of various fish. Furthermore, it improves the process speed and realizesreal-time simulation for movement of fish on common PC.(2) For solving those problems in deformation objects collision detection, such as highcalculation complexity and poor real-time process, this paper proposes a collision detectionalgorithm based on particle swarm optimization (PSO) and GPU. First, realizes the GPUPSOalgorithm based on GPU acceleration. Second, Transforms the problem that computes thedistance between two 3D objects to the optimization problem that computes the nearesttriangle pairs between two group of triangles, Finally calculates the latter problem is solvedby GPUPSO. Compared with the conventional hierarchical bound volume methods, thisalgorithm is faster, more accurate and admits to adjust parameter to balance the relationshipbetween the speed and precision of collision detection.(3)For solving those problems in the traditional fish skin texture process methods, suchas much handwork, low automation and poor ability to be apt to the color change of fish skin,this paper proposes an image color transfer method based on GPU. At first, we realize theimage color transfer method based on GPU, then we transfer this method to the creationprocess of fish skin texture, take invariable texture structure and change color to createcolorful fish skin texture. Compared with conventional texture process methods, thisalgorithm is faster and has better effects. At the same time, the completely automatic processlays a base for the future implementation that fish skin color changes with the environmentand the mood.