Advances in Multimodal Tracking of Driver Distraction - Digital Signal Processing for In-Vehicle Systems and Safety - page 246

Fig. 18.5 Perceived distraction levels based on subjective evaluations. The figure shows the
means and standard deviations for each task across drivers and evaluators

evaluation (3 videos 480). Nine students participated

in the subjective experiment. They assessed only 160 videos (1 video

8conditions 20 drivers ¼

8conditions

160). We read the adopted definition of distraction (Sect. 18.1 )before
the evaluation to unify their understanding of distraction. The presentation of the
videos was randomized to avoid biases. With this setup, each video was rated by three
independent evaluators.
Figure 18.5 gives the means and standard deviations of the perceived distraction
level across secondary tasks. The results indicate that the tasks GPS Following and
Phone Talking are not perceived as distracting as the tasks GPS Operating and Phone
Operating , respectively. Notice that talking on a cell phone increases the cognitive
load of the drivers. Studies have reported that the use of cell phone affect the driver
performance (e.g., missing traffic lights, fail to recognize billboard) [ 20 , 29 ].
This subjective evaluation does not seem to capture this type of distraction. Likewise,
Radio and Picture are perceived as distractive tasks.

20 drivers ¼

18.5 Analysis of Multimodal Features

Our next research problem is to identify multimodal features that can characterize
inattentive drivers. As mentioned in Sect. 18.3 , the corpus includes CAN-bus
information, videos, and audio. Identifying informative features from these nonin-
vasive modalities is an important step toward detecting distracted drivers.

CAN-Bus: One important source of information is provided by the CAN-bus,
which includes steering wheel angle, brake value, vehicle speed, and acceleration.
The car has also sensors to measure and record the brake and gas pedal pressures.