Shadow Volumes

Over the last 10 years, significant progress has been made in the field of computer graphics, especially in real-time rendering. Most noteworthy, the use of dedicated graphics accelerator hardware has found its way from the professional to the consumer market, and their ever increasing power has allowed for rendering almost photorealistic virtual environments while still maintaining interactive framerates. Despite this rapid development, a significant element of computer graphics has been neglected for a long time. Shadows do not only provide more realistic-looking scenes, but also aid the viewer in perceiving spatial relationships. However, due to the additional computational requirements, it has been impossible for graphics accelerators to render shadows and keep framerates high enough to maintain the feeling of immersion for a long time. Although there are several approaches to realize shadows, dynamic environments with a multitude of light sources and complex objects still make high demands on the hardware. In this thesis, a technique is presented to improve the performance of shadow volumes in complex scenes that consist of a large number of individual objects. Several optimization techniques have already been proposed that target applications where the rasterization of the shadow volume polygons is the main bottleneck. However, these optimizations usually assume that the number of individual objects in the scene is rather small compared to the number of geometric primitives (triangles). In such scenes, calculations can be accelerated by using low-polygon approximations of the actual geometry. Most of these existing optimization techniques relieve the graphics hardware at the cost of increased CPU load. If the CPU is already at peak load, these techniques do not achieve any performance gain, but rather worsen the bottleneck in the CPU stage, resulting in an even lower performance. This thesis first presents an overview of current state-of-the-art shadowing techniques that are based on standard shadow volumes. Then, we will try to adapt parts of these techniques to work in complex scenes. Specifically, we will improve visibility determination for shadow volume culling in GPU-demanding scenes with lots of individual objects, as well as present a method for the fast creation of segmented (clamped) shadow volumes that tightly fit the shadow-receiving geometry using vertex programs (shaders).