Interactive Rendering for Large City Models.

摘要

Recent acquisition systems, such as the one developed at the Video and Image Processing Lab at Berkeley, are capable of collecting large, detailed, highly textured models that standard levels of detail (LOD) rendering techniques (18) cannot handle efficiently. We propose an out-of-core rendering engine which applies the cost and benefit approach of the Adaptive Display algorithm by Funkhouser and Sequin to Hierarchical Levels of Detail (HLODs). Unlike the Adaptive Display algorithm, we do not skip objects to maintain interactivity when many objects are visible. Funkhouser and Sequin apply hysteresis by adding a penalty in the benefit heuristics to discourage disturbing visual effects due to fast switching of detail in the model. However, this penalty may not be sufficient if the user is moving around rapidly in the scene. Instead, we have developed a more robust temporal hysteresis by retaining how much detail is rendered over a time period. We have implemented our rendering engine to run on a common personal computer with a standard graphics card. The engine is capable of visualizing, in both walkthrough and fly-through mode, a detailed model of 25 city blocks comprised of 8.9 million triangles and 937 million color pixels. Our engine maintains a constant frame rate while limiting excessive flickering.