Driver Status Identification from Driving Behavior Signals - Digital Signal Processing for In-Vehicle Systems and Safety - page 34

Fig. 3.3 Histograms of the driving behavior signals from highway (top) and city (bottom) traffic
In this section, we discuss feature extraction, driver identification, and driver
behavior signal prediction, and their role for driver behavior modeling. Driver
identification is based on recognition of driving feature vectors using a statistical
model. Our model is designed using a training and test procedure. In the training
part, our algorithm learns the statistical nature of the data from a training set
constructed by extracting the driver behavior features. In the testing part, the
algorithm's accuracy is measured on a testing set, which is completely different
from a training set.
3.3.1 Feature Extraction
A preprocessing step, which is the high-pass filtering of the driving signals including
gas pedal pressure, brake pedal pressure, and vehicle velocity, is applied to remove the
DC component. Then, we apply cepstral analysis, which is a known source/filter