Linear Model Analysis of Real-Time Rendering - Real-Time Rendering: Computer Graphics with Control Engineering - page 49

investigation leads to the conclusion that the summation of separate rendering pro-
cess outputs (frame rates) is equivalent to the output of a single system model using
the combined inputs. In terms of research significance, an accurate system model
can be built upon by the concatenation of separate constituent rendering processes.
This is particularly useful when devising a system model for a complicated render-
ing process is difficult. Furthermore, this principle provides a user with additional
flexibility to manipulate application rendering at a higher resolution.
3.8.5 a dditional n otes
With reference to Equations (3.18) and (3.19), it is important to note that the math-
ematical representation of the system model f is not unique even though it produces
the same stable state output given a same set of input and rendering states. This
is because the dynamics of the rendering system will vary at different operating
(input and output) ranges.
3.9 CONCLUSION
In this chapter, we demonstrated in a systematic manner how our proposed
data-driven modelling framework can produce accurate linear models of real-time
rendering for a variety of applications. We illustrated the extensibility of our frame-
work to handle multiple inputs and validated the accuracy of the resultant model.
More importantly, we validated the case by which the models produced by such
a framework are ultimately useful in the context of interactive rendering with the
introduction of a controller. Finally, our control system is able to eliminate the frame
rate oscillation problem found in typical reactive scheduling frameworks.
Our framework is designed to work on polygonal-based rendering pipelines found
in commodity graphics hardware and it leverages geometry subdivision as a primary
basis for process modelling. As a future research endeavour, we will try to expand
the scope of the model variables for various types of rendering processes wherever
appropriate and possible.
At this juncture, our work is largely based on the subdivision of a single large
mesh. This is useful for applications involving a single object of interest such as
massive model rendering and computer-aided design. However it can be extended to
support multiple progressive meshes in more elaborate applications such as games.