LOD visibility culling and occluder synthesis

Abstract

Level-of-detail occlusion culling is a novel approach to the management of occluders that can be easily integrated into most current visibility culling algorithms. The main contribution of this paper is an algorithm that automatically generates sets of densely overlapping boxes with enhanced occlusion properties from non-convex subsets. We call this method occluder synthesis because it is not sensitive to the way the objects are tesselated but to the space enclosed by them. The extension of this technique by allowing a bounded amount of image error is also discussed. We show that visibility computations can be based on a multiresolution model which provides several representations of these occluders with varying visibility accuracy. Our tests show that occlusion performance in tesselated scenes is improved severely even if no imageerror is allowed.

Inner-Cover of Non-Convex Shapes

“…Andújar et al [1] have presented an algorithm that for a given object O finds a set of inner convex objects. This algorithm generates only axis-aligned bounding boxes.…”

Hardware-accelerated from-region visibility using a dual ray space

“…Synthesis of large convex occluders was studied by Andujar et al [1], and similar techniques can be applied to synthesize 2.5D occluders.…”

On Finding Large Empty Convex Bodies in 3D Scenes of Polygonal Models

“…Rendering these datasets at interactive rates is a challenge in itself. Many methods have been presented to improve rendering speeds, including simple octree-based constructions, mesh simplification, techniques exploiting currently available programmable graphics hardware [35,31,33], and level-of-detail techniques and occlusion culling [2]. Occlusion culling techniques often exploit bounding volumes and hardware acceleration techniques [27,16,7,21].…”

“…Most common visibility culling algorithms include view frustum culling and back face culling [14]. Many visibility methods are based on existence of large occluders to determine a potentially visible set (PVS) [27,2,15,10]. A PVS helps to avoid submitting occluded portions of the environment to the rendering pipeline.…”

“…Furthermore, we are motivated to compute small sets of large convex occluders, as it is often the case that allowing more than one convex occluder can dramatically improve our ability to use them as an effective inner approximation to a shape. Recently, researchers have found out ways to design such a set of occluders by approximating a given shape using axisaligned bounding boxes [2,29].…”