A Multilevel Terrain Rendering Method Based On Dynamic Stitching Strips

Terrain model is an essential part of the virtual geographical environment,which has a wide application in Digital Earth,flight simulation,and computer games.The rendering effect of terrain model has a great influence on the reality of geographical scenes.How to keep the balance of visual quality and rendering efficiency is an important issue for many terrain rendering methods.Early terrain rendering methods focus on reducing the data transferred into GPU as far as possible,and the work of model constructing and updating is completed in CPU totally,which means heavy computation load.Nowadays,with the great improvement of graphics hardware and the development of programmable rendering pipeline,new GPU properties have injected new vitality into the terrain rendering study.Compared with CPU-based methods,the powerful performance of modern GPU makes it possible for rendering millions of triangles on the fly.When rendering large-scale terrain data,it is important to use Level of Detail(LOD)algorithms to create multilevel terrain model.But due to the inconsistency between different LOD levels,visual popping artefacts appear when viewpoint position and direction changes and seriously affect the reality of terrain scene.To eliminate the visual popping artifacts,a real time terrain rendering method based on GPU tessellation is proposed.With the changing of viewpoint condition,the whole terrain is divided into a quadtree of main blocks(M-block),and each of them is tessellated in the GPU tessellation stage.Dynamic Stitching Strips(DSS)are generated concurrently in real time to fill the gaps between adjacent blocks in a seamless manner.The main research contents and conclusions are as follow:(1)Designing multilevel terrain model based on DSS.The terrain model is divided using quadtree and each quadtree node corresponds with a M-block.After the division of terrain,M-blocks are separated by gaps.To fill these gaps and enable the continuous transition of different levels of the terrain model,the DSS structure is introduced,which is generated in real time without complex preprocessing stage.(2)Designing view-dependent multilevel terrain model updating scheme.The height difference between coarse models at different levels and the original terrain model is calculated as the object-space error.According to the ratio relationship in the view frustum and considering the influence of viewpoint position and direction,the object-space error is converted into screen-space error.The updating scheme contains two main stages:updating of quadtree level and recalculating of tessellation levels of M-blocks.By comparing the screen-space error threshold and the screen-space error of the current M-block,the quadtree level is updated and the inner tessellation level of M-block is determined.Besides,to improve the efficiency of our method,a hierarchical view frustum culling scheme is designed which can reduce the updating task significantly.(3)Studying the construction algorithm of DSS.The adjacent relationship of DSS and M-block is recorded in the multilevel updating stage.To guarantee a seamless rendering,the vertices on the common edges of DSSs and M-blocks are generated by using the same tessellation manner as the M-block.These vertices are triangulated into a seamless mesh considering the regularity of their distribution.Experiments demonstrate that our method can render large scale terrain data with a high frame rate.As the viewpoint moving,smooth transition of different levels is achieved.Compared with other method,the visual popping artefacts are avoided and the reality of terrain model is improved.Besides,our method can meet the demand of real-time interactive browsing.