![](bg3.png)
Especially for the calculation of the error in the car’s
position, the ideal position that the driver had to
follow was the right lane of the circuit that
corresponded to constant y position in the Cartesian
coordinates space. Based on this, the error that the
driver made was calculated as the difference from
the ideal position. The error data was normalized for
each driver separately to eliminate as much as
possible the interference of each person’s driving
capacity and style. In this way, the final available
data for analysis were vectors containing for each
driver the normalized error in the car’s position.
The polysomnographical hardware used consists
of a portable medical apparatus capable of acquiring
all the necessary signals. The analysis of these
signals determines the driver’s status. In total, these
signals are 37 and are presented in Table 1. The
software for this polysomnographical acquisition is
the Madcare’s Somnological Studio. The software is
capable of saving all the acquisition session data in
one and only European Data Format file (.edf),
which then is converted into a simple ASCII text
file, using the NeuroTraces edfAsc program. These
text files are loaded and examined in MATLAB. The
sampling rate frequency is set to 200 Hz.
The polysomnographical and mechanical
platform signals are automatically filtered, while the
signals from the steering wheel are first filtered by
the acquisition system, using hardware, and then by
software because of their specific needs. In
particular, for the HRV signal the cut off frequency
has been set to 10 Hz, for the GSR to 0.1 Hz and for
the THE to 0.8 Hz. For determining the correct cut-
off frequencies for every signal, medical advises has
been followed and Fourier analysis has been made.
Steering wheel signals data storing is made using
MATLAB data acquisition files (.daq). These files
are easily handled by MATLAB and also allow
storing the exact acquisition start time and date. All
the data is stored in one matrix, where every column
array corresponds to one sensor and every row array
corresponds to one sampling session (1/200 sec.).
The data from the mechanical platform is stored in
simple text files.
4 PROTOCOL FOR THE
SIMULATION SYSTEM
The simulations are made on two different driver
conditions. In the first part, the driver has slept
during the last night, while in the second he/ she has
been awake for twenty-four hours. In the first state
the nominal conditions of the person are evaluated,
while in the second the altered ones. During the tests
made with the driver not having slept, when sleep is
detected while he/she is undertaking the simulation,
the driver is waken up. In this way, the transition
phases are better examined. The simulations are
always made in dark and noiseless conditions in
order for the person to have much more possibilities
to fall asleep or to lose attention.
Before starting the data acquisition, a
questionnaire is completed by the person responsible
for the simulation, on which the date, the time and
environmental conditions are written. The car at the
start of every simulation session is always positioned
at the same point of the virtual circuit. Each subject,
before driving on the simulation for the first time is
also trained to use the simulator and to always
follow the same pre-defined route.
After these initial procedures, the driver starts the
simulation and the data acquisition is also initialized.
During the procedure and in pre-defined times that
the subject does not know, an obstacle appears on
the screen and the driver has to push the brake. In
this way, his/her reaction time is measured and
stored among all the other parameters acquired. This
response time along with the data from the
polysomnography signals (Rovetta, 1997, Pinelli
1998) determine his/her attention level.
5 OFF-LINE ANALYSIS.
STATISTICS ON THE
ACQUIRED DATA
At the end of every simulation data from both the
sleepy subjects and the control group is divided in
three categories, as shown in table 2.
The purpose of the statistical analysis is to find a
relation between all the measured parameters and the
driver’s vigilance level decrease. The index of the
driver’s vigilance is measured by studying the EEG
signals and the driver’s reaction time to the
appearing obstacles. These analyses focus on two
different directions. First, the general behaviour of
the signals as the driver moves towards sleepiness is
studied. Then, the behaviour of the same signals the
exact minute before a sleep-attack is studied. The
exact time of a sleep-attack is determined, using
EEG Power Spectral Density (PSD) analysis and
medical experience.
For the EEG PSD analysis, after studying all the
possible solutions, the α+β (Eoh, 2005) cerebral
waves method is chosen as the most appropriate
BIODEVICES 2008 - International Conference on Biomedical Electronics and Devices
312