Research On Key Technology Of Modeling And Rendering For Virtual Natural Scene

With the virtual reality applications widely used in many fields, such as militarysimulation, aerospace, natural scenery display, distance education, virtual training and so on,modeling and rendering of virtual natural scene has been a hot research topic. To serve theraising demand of human for realism and real-time, more scene geometry becomes complex,even the most advanced rendering hardware cannot provide interactive rates. High-fidelitymodeling and real-time rendering for virtual natural scene play an important role in manyapplication fields. Therefore, the researches for modeling and rendering of virtual naturalscene have profound theoretical significance and broad application prospect. This papermainly focus on several key techniques, such as mesh simplification, fast terrain rendering,simulation of water waves dynamics and collision detection. The major contents aresummarized as follows:（1）Mesh simplification is a very important research in many domains, our primaryresearch is feature retained mesh simplification algrotithm. This paper present a novel meshsimplification algorithm based on area partition and multi-feature weighted. The proposedmethod consists of two phases: multi-feature weighted optimization and mesh simplification.First, building a hierarchical structure using area partition, then computing the meshmulti-feature saliency metric combining geometric featrue metric and visual featrue metric.Second, the mesh simplification is running with area partition and multi-feature saliencymetric, to get fast and effective siplified models. To evaluate the quality of our algorithm,many existing simplification algorithms are used to compare the quality and speed of meshsimplification.（2）Terrain rendering is a major factor in virtual natrual scene simulation. Interactiverendering of massive terrains still remains a challenging problem because of ever increasingterrain scale and complexity, even the current advanceed hardware design trend cannotprovide fast data access. Our primary point is accelerating terrain rendering with developingdata management techniques of out-of-core and many threads data scheduling in-core, such asview-point position, direction and motion vector of the viewpoin, to solve the above problem.The proposed method uses an off-line process to manage terrain data with techniques such as data block partition, multi-resolution, compression scheme with vertex correlation for everyterrain block. At runtime, the prepatching approach and multi-thread parallel data schedulingare used to reduce the cost of the data exchange and improve the efficienty of renderinglarge-scale terrain.（3）Dynamic ocean surface simulation is also a crucial topic in virtual natural scenesimulation that affects the realism of the applifications. An effective dynamic wave simulationapproach is proposed to solve the problems of realism and real-time. The dynamic wavemodel is divided into fundamental wave, high-frequency detail wave and optical effect. Firstof all, the fundamental wave is implemented by combining cosine wave superposition withocean wave spectrum. Secondly, the high-frequency detail wave is generated using the fractalalgorithm and wave feature analyzing, established a new parameter equation feature-based forfractal parameter optimization. Last but not least, the realistic effects with texture mappingand illumination model is achieved on the GPU.（4）Fast and effective collision detection algorithm is still hard to fulfill in computergraphics. To speedup collision detection, a novel parallel algorithm for collision detectionbased on CPU and GPU is proposed. Firstly, constructing bounding volume and boundingvolume hierarchy for models according to their topology graph to reduce the amount ofbounding volume intersection, and assigning the task of updating bounding volume andbounding volume hierarchy into many-core CPU to parallel computing, with nodeindependence of topology graph. Secondly, the triangles intersection is computing in manyprocessors of GPU, and using spatial Morton code to order geometric primitives as the linearordering to reduce throughput between CPU and GPU. Our algorithm has efficient speedup toconstruct hierarchies of models with up to several million triangles, and is fast for collisiondetection.