Study On Acceleration Technique Of Rendering For Large-scale Terrain Environments

Real-time walkthrough and realistic rendering of large-scale terrain environments is an important topic in computer graphics and virtual reality technique. A terrain surface model composed of height field data provides the observer the primary perception of outdoor scene, so the efficiency of height field rendering is the key factor that determines if it is possible to render the scene in real time. As the model contains huge amount of original data, sometimes even exceeds the capacity of system memory and can not be wholly loaded into the memory, real-time rendering confronts with many difficulties. Even though the top graphics card has become more and more powerful so that over one million of triangles can be rendered per second, interactive terrain rendering using brute-force approach is still impossible, and acceleration technique should be developed to speed up the rendering of scene. We carry on our research work to this problem, try to find the optimal acceleration method or combination of acceleration techniques to further improve the rendering frame rate while still maintaining the image fidelity unchanged. For those applications based on walkthrough of terrain environments such as VR GIS, virtual battle field, 3D game or driving simulation, we investigate the work and concentrate on visibility culling of scene with other accessorial acceleration methods such as LOD, optimal data paging, hardware-assisted speedup etc.The main contributions of this thesis are: This thesis proposes a conservative hierarchical visibility computation method for terrain environments, which belongs to from-region visibility problem. The combination of hierarchical visibility with multi-resolution model leads to efficiently rendering large-scale terrain in real time with continuous level of detail. Hierarchical visibility is computed offline in preprocessing stage, which needs less storage andcomputation overhead than previous algorithm. Visibility information is used to cull those invisible blocks of terrain at run time, thus reduces the time spending on tessellation and rendering. The visibility determination is conservative and can avoid visual artifact induced by improper culling. The view-dependent simplification algorithm based on quadtree structure can also be used to render large-scale point model, we implement a hybrid rendering approach that combines the points in parameter space with polygons in geometric space.This thesis proposes a cascading occlusion culling algorithm based on horizon. From-region visibility algorithm has its inherent disadvantages because it may cost much storage and time on preprocessing. The contradiction becomes obvious for out-of-core dataset. Outdoor terrain has special geographical feature information, which we extract from the original dataset offline and make use of them to construct occlusion horizon at run time, and temporal coherence is well applied while performing cascading occlusion culling between continuous frames. We use both view parameter and visibility information as the standard of LOD selection, and we propose a visibility-driven CLOD framework for terrain. Besides occlusion culling algorithm in object space, we also implement hardware-assisted visibility culling in image space using extracted features.This thesis proposes a new error metric of simplification that well suits walkthrough application and proposes a scheme that uses incremental horizon to control data loading and geometry rendering for out-of-core terrain dataset. Out-of-core model has more difficulty than general large-scale scene in real-time rendering, so special treatment is required. The error metric proposed in this thesis is silhouette-preserving and shading-preserving. It accords with the visual perception of human's eye when walking through. The model simplification using this metric can further reduce the geometric complexity while maintaining image quality. We use potential silhouette as the occluder to construct incremental horizon, control the data exchange between system memory and secondary storage,...