Introduction - Real-Time Rendering: Computer Graphics with Control Engineering - page 3

1. We describe a novel framework by which the real-time rendering pro-
cess may be modelled accurately. This framework involves the adoption
of data-driven system identification methodology. Previous attempts to
characterise the rendering process via only observable variables and
case-specific formulations led to inaccurate models. Our model addresses
these shortcomings.
2. Apart from linear models, our data-driven framework is extended to non-
linear models using soft computing techniques such as neural networks and
fuzzy models.
3. We developed control system frameworks for both linear and non-linear
models in real-time rendering using (a) PID control with and without gain
scheduling and (b) fuzzy control with and without adaptive neural networks.
The application of our control frameworks has shown much better resource utilisa-
tion in the real-time rendering process than earlier work that generally demonstrated
coarse performance tracking.
1.3 SCOPE OF WORK
Real-time rendering is a vast topic in the field of computer graphics. Although the
modelling techniques and control framework may be applicable to areas such as
volume- and image-based rendering, our study deals with polygonal-based rendering
pipelines found in commodity graphics hardware and it leverages geometry sub-
division technique as a basis for controlling the input to the rendering system.
At this juncture, our work is based largely on the rendering of a single large
3D mesh that is used as a pseudo-representation of more complex 3D scenes with
numerous objects. From a different perspective, this system is useful for applications
involving a single large object of interest, for example, massive model rendering and
computer-aided design.
Since the focus of this research is on real-time rendering relating to the response
time of a system in an interactive environment, we consider the time required to
render an image (frame) as the critical performance metric. While computer graphics
activity is essentially visual, the quality of the generated image is frequently taken as
the next most important metric for assessment. However, due to the subjectivity and
complexity involved in processing image comparisons, the image quality component
is omitted as a performance object in this work. From the system perspective, the
real-time rendering framework proposed in this research is flexible to accommodate
a multiple-input-multiple-output (MIMO) configuration. This means the user has
the full freedom to implement additional output variables, which may include image
quality related performance variables.
1.4 BOOK OUTLINE
Chapter 1 provides the background and motivation that led to this research.
Chapter 2 discusses the fundamental knowledge in two key disciplines related to
this research—real-time computer graphics rendering and system identification