Visibility Computations in Densely Occluded Polyhedral Environments

• Authors: Seth J Teller
• Venue: Visibility Computations in Densely Occluded Polyhedral Environments
• Year: 1992

This thesis investigates the extent to which precomputation and storage of visibility information can be utilized to accelerate on-line culling and rendering during an interactive visual simulation of a densely occluded geometric model. We have shown that practical visibility preprocessing and on-line culling is achievable, both theoretically and as a functioning implementation, for an important class of densely-occluded environments (e.g., architectural models). The visibility scheme first subdivides a geometric model into spatial cells, introducing cell boundaries wherever major occluders are present in the model. Next, a coarse visibility determination links cells, and objects within them, that are mutually visible through sightlines, or incident on conservative light bundles. This coarse visibility determination constitutes a per-cell potentially visible set or PVS of objects as an upper bound on the set of objects visible to any actual observer in a given cell. Thus, a simulated actual observer can be tracked through the spatial cells, while the coarse visibility information stored with each cell is retrieved, and subjected to on-line culling operations. The observer''s instantaneous view position and field of view are used to reduce the set of objects potentially visible from {\em anywhere} in the cell, to those potentially visible from the observer''s eyepoint. The resulting subset, typically a small fraction of the model data, is then issued to graphics hardware for rendering and discretized hidden-surface removal. These subdivision and visibility determination techniques have been implemented for general polyhedral geometric models in three dimensions. Using a furnished model of a planned computer science building as test data, we have achieved static culling rates of over 90\% and dynamic culling rates of more than 99\%, on average, decreasing the average refresh rates of a simulated walk through the model by a factor of about one hundred. The subdivision and visibility precomputation stages required several hours of compute-time for a five-story building, a reasonable figure given the design time cycle for a project of this size. The results of the visibility computation incurred less than 10\% storage overhead (15 megabytes) with respect to the amount of storage required by the geometric model itself (180 megabytes).