Dynamic scene occlusion culling in architectural scenes based on dynamic bounding volume

Abstract

Visibility algorithms have recently regained attention because of the ever increasing size of polygon datasets and more dynamic objects in a scene. Dynamic objects handling makes large polygon datasets impossible to display in real time with conventional approaches. Therefore, occlusion culling techniques are required for output-sensitive rendering. Most scenes are displayed with static objects and only a few use dynamic objects in their visualization. In this thesis, the aim of the research carried out, is to handle dynamic objects efficiently with faster frame rate display. This algorithm is implemented using portal occlusion culling and kD-tree, which is suitable for indoor and architectural scenes. An occlusion culling technique was developed for handling dynamic objects in static scenes using dynamic bounding volume. Dynamic objects are wrapped in bounding volumes and then inserted into spatial hierarchical data structure as a volume to avoid updating the structure of every dynamic object at each frame. Dynamic bounding volumes are created for each occluded dynamic object by using physical constraints of that object and are assigned with a validity period. These bounding volumes and validity periods are later inserted into kD-tree. The dynamic objects are ignored until the bounding volume is visible or the validity period has expired. After numerous tests and analysis have been done, dynamic bounding volume culling shows better performance than portal culling especially when there are many low speed dynamic objects in the scene. Dynamic bounding volume culling proved to be efficient in avoiding enormous calculations of dynamic object’s position thus improves the rendering speed.