Distributed visibility servers

Abstract

This thesis describes techniques for computing conservative visibility exploiting viewpoint prediction, spatial coherence and remote visibility servers to increase the rendering performance of a walk through client. Identifying visible (or partially visible) geometry from an instantaneous viewpoint of a 3-D computer graphics model in real-time is an important problem in interactive computer graphics. Since rendering is an expensive process (due to transformations, lighting and scan-conversion), successfully identifying the exact set of visible geometry before rendering increases the frame-rate of real-time applications. However, computing this exact set is computationally intensive and prohibitive in real-time for large models. For many densely occluded environments that contain a small number of large occluding objects (such as buildings, billboards and houses), efficient conservative visibility algorithms have been developed to identify a set of occluded objects in real-time. These algorithms are conservative since they do not identify the exact set of occluded geometry. While visibility algorithms that identify occluded geometry are useful in increasing the frame-rate of interactive applications, previous techniques have not attempted to utilize a set of workstations connected via a local area network as an external compute resource. We demonstrated a configuration with one local viewer and two remote servers.