Dynamic Irregular Octrees

As consumer graphics hardware becomes faster and more capable, it is becoming desirable for realtime interactive content to become more dynamic. However, the static spatial partitioning mechanisms used in current interactive systems, such as games and other virtual reality environments, are not suitable for a fully-dynamic interac- tive environment. In this paper, we take a new look at octrees, and create a dynamic octree-like data structure which allows for real- time modification of a scene while retaining many of the benefits of a static spatial partitioning. Additionally, we explore a mechanism for providing fully-dynamic occlusion culling using this dynamic spatial partitioning. nications pipeline with the graphics processor, the CPU can spend more time on per-frame visibility determination without degrading the overall performance. As such, an algorithm which provides an extremely-conservative working set is acceptable as long as there isn't a significant penalty to maintaining the partitioning or reject- ing the extraneous objects. In this paper, we describe a fully-dynamic adaptation of one of the simplest and fastest spatial partitioning algorithms - the octree - and show that even though it is extremely conservative, it still boosts performance by an order of magnitude over a brute-force test. We then adapt a simple but effective occlusion-culling algo- rithm from a static octree to a dynamic structure, providing a fast, output-sensitive, and totally-dynamic visibility algorithm with no need for constraints of any sort. Finally, we show how to use the dynamic octree with a hierarchally-modeled scene, and examine the use of the dynamic octree for collision detection.