Driver Status Identification from Driving Behavior Signals - Digital Signal Processing for In-Vehicle Systems and Safety - page 30

different cognitive stress/distraction conditions using driving behavior signals.
Our objective is to search out and examine the effects of cognitive distraction
conditions on driving behavior and inquire whether driving behavior signals are
characteristic information for every driver. We investigate task identification
performances, where our earlier findings are presented in [ 17 ].
In this chapter, we present our contributions on the following three major
problems:
• Driver Identification : Identification of a driver using behavioral signals is one
the most interesting in-vehicle signal-processing problems. In this study, we use
driving behavior signals such as vehicle speed, gas pedal pressure, brake pedal
pressure, and distance from the vehicle in front for driver identification. First, we
investigate the characteristics of these signals and present a selected set of
driving statistics. Then we define a statistical driver identification system and
evaluate this system experimentally.
• Driver Status Identification : Distractive conditions cause important safety
problems to drivers. Studies have shown that nearly 80% of traffic accidents
occur due to driver inattention, which are commonly results of distractive
conditions. Navigation systems and other services in vehicles introduce many
secondary driving tasks that can increase accident risk. Thus, developing a distrac-
tion detection method would be very beneficial for in-vehicle system to reduce the
effects of distraction. In this study, driving experiments were done under some
distractive conditions, which can be considered as the secondary driving tasks
stated above. These tasks are dialog on cell phone, including route navigation and
online banking, conversation with passenger on-board, and signboard and license
plate reading. We investigate the statistical nature of driving behavior signals under
different driving tasks, which are defined as distractive conditions. Then we
attempt to detect distractive conditions using statistical classifiers.
• Driver Behavior Prediction : Human factors play a big role in traffic accidents.
Predicting driving behavior is an important issue since it has a significant effect
on decreasing human-caused accidents. Drivers' behavior is strongly related to
their past actions, so in this study, we construct a driver behavior prediction
model using drivers' past behavior signals. The driver behavior prediction model
consists of temporal clustering with hidden Markov models (HMM) and mini-
mum mean-square error (MMSE) estimation within each temporal segment. We
also investigate the influence of road conditions and distractive conditions on our
prediction model.
3.2 Driving Behavior Signal Characteristics
Driving signals differ in how and under which conditions the driver use vehicle
control units, such as pedals, driving wheel, etc. We aim to model individual
differences among the selected drivers and identify the drivers by using gas