An Adaptive Hierarchical Occlusion Culling Algorithm for Interactive Large Model Visualization Abstract

This paper describes a new occlusion culling technique for efficient visibility determination in real-time environments. There are two major contributions of our work. First, we present an efficient algorithm to determine a set of potential occluders by sampling the virtual environment by sending rays from the camera into the view frustum. Thus, the set of potential occluders is always a subset of actually visible objects. We show how we can reduce the number of occlusion tests based on the fact that we use only visible occluders. Second, we present an algorithm that reduces the number of unsuccessful tests for occlusion significantly. We control the activation and de-activation of potential occluders for actual testing based on previous testing results. This control reduces the number of occlusion tests by a factor of 1.5-10 depending on the underlying scene geometry compared to testing against all potential occluders.This results in a significant increase of the frame rate. Our algorithms do not make any assumptions on how the actual occlusion test for a single object against a different object is implemented. Rather, we present a high level frame-work to control the occlusion tests. We exploit frame-to-frame coherence and the underlying techniques can easily be adapted to dynamic environments with moving objects. 1