Driver Status Identification from Driving Behavior Signals - Digital Signal Processing for In-Vehicle Systems and Safety - page 31

pedal pressure, brake pedal pressure, vehicle velocity, and fusion of these signals.
We also benefit from car-following distances. Driving behavior characteristics
differ from person-to-person under different distractive conditions. In order to
examine the effects of these distractive conditions, we investigate how driving
behavior signals differ across driving tasks. Statistically significant clues of this
investigation are used to define a driving status model. This section presents general
characteristics and statistics of the driving behavior signals from the UYANIK
database, feature representation of these driving behavior signals, and the statistical
clustering, identification framework for the driver and driving status and predicting
driver behavior.
3.2.1 Data Collection
Driving behavior data was supplied by the Drive-Safe Consortium in Turkey with
the test vehicle, UYANIK, which is a sedan car equipped with various sensors. The
UYANIK database includes synchronous audio-visual recordings, CAN-Bus
readings, pedal-sensor recordings, 180 laser range finder, and XYZ accelerometer
recordings [ 13 ].
The data collection route is around 25 km at about 40 min, starting and ending at
the OTAM Research Center in the ITU Campus in Ayazaga. It consists of two
1.5 km-very-busy city sections, followed by the TEM highway with much less
traffic. Next, the route goes through the city, and then it goes back to the OTAM at
ITU campus. The last segment is very busy with local traffic. The route is the same
for all drivers. However, road conditions may differ depending on traffic jam and
weather in Istanbul. We use a subset of the UYANIK database including driving
behavior signal recording sessions of 20 male and 3 female drivers.
There are four primary tasks in the UYANIK database: (1) reference driving
which includes no specific driving task, (2) dialog on cell phone which includes
online banking application and navigational dialog, (3) signboard reading in which
driver reads road-by signs and license plates aloud, and (4) dialog with passenger
where driver talks with the on-board passenger.
3.2.2 Driving Behavior Signals
We consider gas and brake pedal pressure signals, velocity from CAN-Bus, and car-
following distance from the laser range finder as driving behavior signals. The gas,
brake, and velocity signals are all sampled at 32 Hz, and the laser range finder
sweeps 180 at every 2 s. Samples of driving behavior signals are given in Fig. 3.1 .
The laser range finder in front of the vehicle records two-dimensional (x, y) data
consisting of horizontal and vertical distances. Figure 3.2 shows the Laser Scan
Reading and the photo for a selected driver recorded at 12:56 PM on April 6, 2007.