Interactive visual realism using distributed rendering

Visual realism is necessary for virtual reality. To be so convincing that the user perceives the virtual environment as real, the scene presented must faithfully model the actual environment the user expects to see; a highly accurate, fully modeled, interactive environment will be seen as truly ""virtually real."" Most contemporary virtual reality systems compose the view of their virtual environments using high-speed dedicated computer graphics hardware. Such hardware is incapable of rendering many of the complex visual features common to natural environments at the interactive rates required by virtual reality. Ray tracing is a more powerful rendering method; unfortunately, pure software implementations on contemporary general purpose hardware are too slow for use in interactive systems.; This dissertation proposes a possible solution to the problem of high-speed visual realism: the combination of the speed of fast graphics hardware with the power of software ray tracing, distributed over several machines linked by a fast network. The result is a hybrid rendering paradigm, capable of high speed and high fidelity. As fully accurate modeling may be too expensive for complex interactive systems, the hybrid system requires adaptations of existing methods as well as new high-speed algorithms for the approximation of highly complex, realistic effects.; This new paradigm is examined in the context of a working rendering system. This system is capable of producing images of higher fidelity than possible through the use of graphics hardware alone, able both to render images at speeds useful for interactive systems and to progressively refine static, high quality snapshots. Experimental results suggest projections about the performance of future systems, encouraging further development to increase the fidelity of future systems. This work increases the feasibility of truly realistic virtual reality.