THE UC BERKELEY SYSTEM FOR INTERACTIVE VISUALIZATION OF LARGE ARCHITECTURAL MODELS

Realistic-looking architectural models with furniture may consist of m illions of polygons and require gigabytes of data--far more than today 's workstations can render at interactive frame rates or store in phys ical memory. We have developed data structures and algorithms for iden tifying a small portion of a large model to load into memory and rende r during each frame of an interactive walkthrough. Our algorithms rely upon an efficient display database that represents a building model a s a set of objects, each of which can be described at multiple levels of detail, and contains an index of spatial cells with precomputed cel l-to-cell and cell-to-object visibility information. As the observer m oves through the model interactively, a real-time visibility algorithm traces sightline beams through transparent cell boundaries to determi ne a small set of objects potentially visible to the observer. An opti mization algorithm dynamically selects a level of detail and rendering algorithm with which to display each potentially visible object to me et a user-specified target frame time. Throughout, memory management a lgorithms predict observer motion and prefetch objects from disk that may become visible during imminent frames. This paper describes an int eractive building walkthrough system that uses these data structures a nd algorithms to maintain interactive frame rates during visualization of very large models. So far, the implementation supports models whos e major occluding surfaces are axis-aligned rectangles (e.g., typical buildings). This system is able to maintain over twenty frames per sec ond with little noticeable detail elision during interactive walkthrou ghs of a building model containing over one million polygons.