Terrain modeling for real-time photo-texture based visual simulation

There are many applications, such as pilot training, mission rehearsal, and hardware-in-the-loop simulation, which require the generation of realistic images of terrain and man-made objects in real-time. One well-known approach to meeting this requirement is to drape photo-texture over a planar polygon model of the terrain. The real time system then computes, for each pixel of the output image, the address in a texture map based on the intersection of the line-of-sight vector with the terrain model. High quality image generation requires that the terrain be modeled with a fine mesh of polygons while hardware costs limit the number of polygons which may be displayed for each scene. The trade-off between these conflicting requirements must be made in real-time because it depends on the changing position and orientation of the pilot's eye point or simulated sensor. The traditional approach is to develop a data base consisting of multiple levels of detail (LOD), and then selecting for display LOD's as a function of range. This approach could lead to both anomalies in the displayed scene and inefficient use of resources. In this paper, a new approach has been developed in which the terrain is modeled with a set of polygons and organized as a graph with each node corresponding to a polygon. This graph is pruned to select the optimum set of nodes for each eye-point position. The novelty of the proposed graph pruning lies in the coherence suggested in the selection of terrain LODs and texture LODs. The error in terrain representation is expressed in terms of the error in the texture pixels. This allows different error criteria to be used for different applications. Texas 76019 As the point of view moves, the visibility of some nodes drops below the limit of perception and may be deleted while new nodes must be added in regions near the eye point. Our pervious work has been generalized to include the case where the refinements of the polygonization are not simply nested and the graph is not a simple tree. This generalization prevents the occurrence of gaps in terrain representation.