Perception-guided simplification for real time navigation of very large-scale terrain environments

Interactive navigation over very large-scale terrain environments is a highly challenging topic since the model involved often contains huge amount of geometries and textures, even exceeds the capacity of system memory. In this paper, we put forward a high-performance technique for real-time walkthrough in very large-scale terrain environments. By the technique, perception-guided simplification is adopted through application of a new error metric-Constrained Normal Cone in the preprocessing stage. Both silhouette-preserving and shading-preserving criteria are satisfied in our navigation system. Furthermore, the pre-extracted Potential Silhouette from each block can be used to construct the incremental horizon at run time for delicate visibility testing. The consequent visibility information then can prevent the unnecessary data from loading into the main memory, so that the workload of out-of-core data paging and the number of triangles rendered are substantially reduced, and the frame rates are highly increased. Experiment results show the high performance of our novel approach.