Advances in Multimodal Tracking of Driver Distraction - Digital Signal Processing for In-Vehicle Systems and Safety - page 239

Fig. 18.1 Monitoring driver behavior system using multimodal information
not surprising since 30% of the time when a car is moving, the driver is involved in
secondary tasks that are potentially distracting [ 3 ]. With the development of new
in-vehicle technologies, these numbers are expected to increase. Therefore, it is
important to identify and develop feasible monitoring systems that are able to detect
and warn inattentive drivers. These systems will play a crucial role in preventing
accidents and increasing the overall safety on the roads.
Distraction can affect visual, cognitive, auditory, psychological, and physical
capabilities of the driver. Distraction is defined by the Australian Road Safety
Board as “the voluntary or involuntary diversion of attention from the primary
driving tasks not related to impairment (from alcohol, drugs, fatigue, or a medical
condition)” [ 4 ]. Under this well-accepted definition, the driver is involved in
additional activities that are not related to the primary driving task, which include
talking to a passenger, focusing on events or objects, and manipulating in-car
technologies. As a result, the driver reduces his/her situational awareness, which
affects his/her decision making, increasing the risk of crashes.
We have been working in detecting inattentive drivers by combining different
modalities including controller area network-bus (CAN-bus) data, video cameras,
and microphones [ 5 , 6 ]. Our long term goal is to develop a multimodal framework
that can quantify the attention level of the driver by using these noninvasive sensors
(Fig. 18.1 ). Instead of relying on simulations, the study is based on recordings with
actual drivers in real-world scenarios using the UTDriver platform - a car equipped
with multiple sensors [ 7 ]. First, we have studied the changes observed in features
across modalities when the driver is involved in common secondary tasks such as
operating navigation systems, radio, and cell phone [ 5 ]. Then, we have proposed a
regression model based on relevant multimodal features that can predict driver
distraction. The results have shown that the outputs of the proposed system corre-
late with human subjective evaluations.
This chapter discusses the state-of-the-art in detecting inattentive drivers
using multiple sensing technologies. It describes previous studies and our own
contributions in the field. Notice that we only focus on distractions produced by
secondary tasks. We do not include distractions or impairments produced by alcohol,
fatigue, or drugs [ 8 , 9 ].