Accelerated Rendering Method Of Large And Complex Scene And Implementation

The existing visibility culling algorithm and level of detail algorithms are based on CPU required spending more CPU resources to calculate the potential visible set and the level of detail model, in order to maximize the reduction in the number of triangle and get the tightest triangle set. Their defects are that the overhead for speeding up is more than the advantage from reduction of the number of triangles. Now hardware rendering has became the industry's standard, and is able to process and render a large amount of triangles per frame. We will integrate and improve some of the existing speeding up rendering algorithm. We will reduce CPU overhead and improve GPU utilization to balance the CPU and GPU overhead, in order to rendering 3d complex scene in interactive and real-time mode.An outdoor urban scene is the typical large-scale and complex scene where there are lots of buildings and these buildings are highly occluded for each other. For the highly occluded scene, occlusion culling is the most effective way to speed up the rendering. We have researched and implemented coherent and hierarchical visibility computing based on GPU occlusion queries. We divide the scene space and build hierarchical space structure accord to its inherent spatial topology. We exploit spatial and temporal coherence of visibility to minimize the number of nodes for visibility test and to reduce the delays due to the latency of query results. Our experiment proves that the occlusion culling method can effectively improve the large-scale and complex scene rendering speed.The number of triangle faces is very large in a geometry model built on terrain high field map and interactive rendering full geometry details of terrain model is beyond the rendering capacity of general graphics hardware. We have researched and implemented multi-hierarchies rendering of large-scale complex terrain which blending view-dependent multi-resolutions LOD and hierarchical visibility computing based on horizon occlusion culling. We have researched and improved the LOD algorithm based on GeoMipMaps which partitions the whole terrain into some blocks to build hierarchy structure of terrain space and generate suitable resolution geometry details for each block according to its visual importance to eyes. The LOD computing is simple and efficient, consumes very little CPU resources. We calculate hierarchical horizon occlusion diagram of the overall terrain in off-line time and use this occlusion diagram to determine the node of terrain hierarchy in real-rendering time. Our experiment proves that the multi-hierarchies rendering method could render large-scale complex terrain scene in interactive and real-time mode.