Modeling And Rendering Of Large Scale Natural Scene

The modeling and rendering of large scale natural scene are important for virtual battle field, 3D interactive game, film and video etc. With the further spread of virtual reality applications, there are more and higher requirements on simulation scale, simulation precision, and performance of interaction. These requirements promote rapid development of computer graphics and graphics hardware.To meet the demands of the applications of virtual battle, this dissertation has studied some key technologies such as simulation of dynamic cloud, navigation of large-scale terrain, and rendering of realistic ocean wave. The main contributions of this dissertation are as follows:This dissertation has studied the LOD model and algorithm of large-scale terrain in depth. With the wide spread and fast development of powerful graphics process unit( GPU), the traditional terrain models and algorithms depending on accurate culling and elaborate multi-resolution LOD layout are no longer good solutions for navigation and rendering of large scale landscape. This dissertation has brought forward a patch-LOD algorithm of terrain rendering, which can take advantage of capacity of GPU, and balance the overload of CPU and GPU. In this dissertation, the Terrain tile pyramid (TTP) and terrain summary pyramid (TSP) is introduced, and the access model of large scale terrain and the scheduler of terrain tiles are implemented based on TTP and TSP. The TTP's summary information is represented by the corresponding TSP, and TSP's storage is much smaller than TTP's, therefore TSP is convenient to be used in view-frustum culling, occlusion culling and control of LOD. Just for stitching the terrain tiles of different level of detail seamlessly, this dissertation introduces the Index Buffer Templet (IBT) and Joint Index Buffer Templet (JIBT). Additionally, this dissertation introduces Top Clamp Plane (TCP) and Bottom Clamp Plane (BCP). The TCP takes the place of occlude terrain tile during computing the visibility of a. terrain tile and the BCP takes the place of occluder terrain tile during constructing the horizon visibility. The TCP and BCP make the occlusion culling more efficient than before. The experimental results show that the Patch-LOD algorithm can satisfy the requirements of the large-scale, interactive virtual scene applications.This dissertation has studied the geometry detail and texture detail of terrain in depth and proposes an algorithm of dynamic systhesis of terrain texture. Due to the limitation of capacity of computer and bandwidth of I/O interface, the geometry detail and texture detail of large scale terrain is restricted in scale and precision. For producing realistic landscape, This dissertation presents the dynamic systhesis of terrain texture based on the terrain algorithm of Patch-LOD , which takes full advantage of mechanism of dynmic scheduling of terrain tiles and the mechanism of LOD. The experimental results show that, this algorithm can produce realistic terrain texture with small cost in computation when there is no according large scale preloaded texture, or the preloaded textures are difficult to be loaded in real-time.This dissertation has studied the modeling of ocean, and proposes a method of simulation of shallow water near the shore. When the ocean wave arrive at the area of shallow water, the wavefronts will bend because of refraction, and diffraction. The forward direction ocean wave will trend to be vertical to the contour of terrain. According to the above analysis of shallow water, this dissertation proposes an algorithm based on decomposed model of direction-frequency spectrum. In this method, the ocean wave will be divided into two components: the directional component is modeled with sine wave, and the random component is modeled with Perlin noise. The experimental results show that this algorithm can provide realistic image and high frame rate.This dissertation has studied simulation of cloud in depth and proposes a two-stages simulation method of dynamic cloud based on Lattice Boltzmann Model (LBM) and impoved Cell Automaton (CA). Due to the physical complication, the simulation method of dynamic cloud based on physical model of cloud and atmosphere is too difficult to be implemented on normal personal computer. To ensure the reality of simulation results and high frame rate of interactive system, this dissertation brings forward a new method: two-stages method of modeling the dynamic cloud. In this method, the simulation of cloud is divided into preprocess and real time process. During preprocess, the wind field of atmosphere is produced. During real time process, the dynamic clouds are produced on condition the complicated wind field is computed in preprocess. The former is based on the lattice Boltzmann method which can embody the properties of atmosphere, and the latter is based on cell automaton which is effient and easy to deal with the boundary condition.Based on the above algorithms and methods, an experimental system is built, in which large scale landscape and seascape and ocean wave near shore are integrated. The results of our experimental system demonstrate that the algorithms and models proposed in this dissertation are effective and efficient.