Linear Model Analysis of Real-Time Rendering - Real-Time Rendering: Computer Graphics with Control Engineering - page 19

3 Linear Model Analysis

of Real-Time Rendering

3.1 INTRODUCTION
The real-time computer graphics rendering process embodies complex state transi-
tions and fast dynamics amidst observable steady-state behaviour. To yield realistic
or visually useful graphical information, the rendering process may be loaded with
myriad combinations of the input variables and states to the point where it is impor-
tant to describe this function in simple terms.
In this chapter, we describe the application of system identification methodology
to real-time rendering. The basis for such an approach is that the rendering process
may be treated from a system perspective as a data processing function. This allows
us to analyse the process input and output data to establish some formal relationship
between them.

3.2 BACKGROUND
The perennial and increasing demands for fidelity, coupled with hardware archi-
tecture advancements, pose many challenges to researchers and developers as they
endeavour to find the optimal solution to accommodate both speed and quality of
rendering. To this end, key techniques developed since the evolution of computer
graphics three decades ago revolve around their ability to reduce the rendering load
at application runtime. They are largely based on the principles of visibility reduc-
tion, geometry decimation, image-based methods, and more recently, techniques
such as programmable shading.
As space does not permit an exhaustive review of these research efforts, we refer
the reader to the comprehensive surveys by Cohen-Or et al. [12], Haines [13], and
Akenine-Moller et al. [14]. Despite the ability of each approach to reduce rendering
loads during runtime, their common weakness lies in the inability to guarantee stable
frame rates.
In this chapter, we introduce a novel framework for obtaining an accurate model
of an interactive rendering process. This framework is based upon the system iden-
tification methodology [1] that is a mature field of study associated with systems and
control theory.
In addition to expanding understanding of the dynamics relating to the rendering
process, the objective of modelling this process is to establish the groundwork for a
control framework. Only with an accurate model can we design this control frame-
work around the rendering process to yield the sustainable performance we desire.

15