Preliminaries - Real-Time Rendering: Computer Graphics with Control Engineering - page 11

Design experiment and
collect data

Y

Data require post-
processing

Post-process data

N

Choose model
structure

Select model
parameters and fit
model to data

Validate model

N Y

End

Model is acceptable

FIGURE 2.5 Process flow in system identification methodology.

preliminary information obtained from the system. The parameters of this model
structure are then computed based on the set of experimental data collected previ-
ously. A portion of this data is allocated for model validation and the entire process
from choosing a model structure to validation is repeated until the user-defined
simulation performance criteria are met.
From a system identification perspective, we treat the real-time rendering process
as the subject to be modelled. Since the rendering process cannot be described intui-
tively by physical laws such as mass, velocity, and temperature, black-box modelling
is adopted. The system is first tested with a set of predefined inputs and the outputs
are collected. This input-output dataset that captures a certain dynamic range of
the behaviour of the system is then used with mathematical regression techniques to
derive the estimated model.
Due to the scope of this topic, we briely summarise the steps in the system iden-
tification process below. A detailed and authoritative coverage of this topic can be
found in Ljung's topic [1].

2.2.1 d ata c ollection
To obtain an effective model of a system, it is necessary for the measured data to
capture and show the behaviour of the system adequately. An appropriate experi-
mental design can ensure that the correct variables and dynamics of the system are