5. Fatigue modeling

All the characteristics calculated during the various operations described above are fed into a numerical model to predict the degree or state of driver fatigue. There are several types of numerical model used in fatigue detection. These include rule-based methods, machine learning and deep machine learning.

The simplest system available is a rule-based system based on one or two parameters, such as PERCLOS and/or FOM (mouth opening frequency). A threshold is used to identify the state of the operator with the PERCLOS and/or FOM value: alert or tired. The system is based on first-order rules such as "if, and, or, then", with numerical comparators such as "greater or lesser".

It has been shown that it is preferable to use several parameters to obtain more reliable results.