Large-scale Complex Scene Rendering Optimazation On GPU

With the rapid development of computer graphic devices,rendering,as the key technology to display kinds of colours,especially in popular area,like virtual reality and game engine,has been paid great attention by more and more researchers.The rendering of large-scale complex scene,is widely used,though it's the most difficult part to improve algorithm effiency.In reality,the demand of the rendering quality kept going higher.Thus the user's hardware'capability is limited,how to balance the contradiction between them,has become the research hotspot and main barrier,with rendering quality guaranteed even promoted.So,with the limited capability of given hardware,how to improve the rendering algorithm efficiency of large-scale complex scenario,will bring important research value and far-reaching research significance to the development of computer graphics.For a long time,in view of the large-scale complex scene rendering optimization methods emerged continously,those which were widely used including occlusion culling algorithm,LOD algorithm,instantiation technology,etc.Based on former researchers' studies,this thesis proposes an innovatice and efficient rendering optimazation algorithm,with the help of Unity3 D rendering engine,respectively for the large-scale terrain and the complex models: the first part,under the allowance of certain amount of error and real-time rendering,established a highly efficient data loading and release strategy between memory and data,based on improved GeoClipmap algorithm.Relieved the contradiction between the limited computational resources and enormously amount of data in the process of rendering,optimized complex terrain rendering effiency;The second part,the author abandoned hardware block queries due to its own asynchrony,choosed Hierarchical Covered Image algorithm,the establishment of Hierarchical Covered Image is the key step in occlusion culling algorithm,during the pretreatment stage,through the preset occlusiong culling conditions,screened the objects in the scene,created a mask database;During the operational procedure,the algorithm extracted a set of objects as obstruction at the current viewport,built hierarchical block diagram and rendered it into image space,finally used hierarchical block diagram of objects to extract every visible objects.At the same time,through an array of visible objects,the hierarchical occlusion algorithm is combined with the improved viewpoint-dependent LOD algorithm.In the pre-processing phase,the OBB tree generates a collection of different granularity units of space,then the visibility of each space unit is calculated for each space unit from different viewpoints.At the operational stage,the algorithm takes the current view point visible and chooses LOD based on the distance between the object and the viewpoint.Finally,the results are written to the triangle rendering sequence,which is sent to the rendering line.Finally,the paper took experiments with kinds of different types of large-scale complex scene rendering respectively,compared with existing technology,made data comparison and the performance analysis with the result of the experiment,verify the validity and effectiveness of the proposed algorithm: in large-scale complex scene rendering,with the quality of rendering guaranteed,algorithm made a significant improvement to the rendering efficiency.