Abstract:
Vehicle systems that use driver behavior data to determine safe and unsafe behavior
need to operate in real time and in real chaotic environments. The research in developing
these systems do not have publicly accessible data sets that would aid in research and
development. In order to create these data sets, experiments and data collections need
to be performed in an unconstrained real world environments or in a highly constrained
environment. This thesis proposes a tool to collect real time data in unconstrained real
world environment, called Live Driving Detection (LiDD). LiDD labels driver head rotations
to determing where the driver is looking and combines the state of the car from the CANBus
network to add more context to the produced data. The labeling process is designed to be
simple in order to quickly label a given instance of data called a frame. LiDD is able to
output labeled driver data fused with vehicle state at approximately 6 HZ. LiDD's utility
was evaluated in three common real world environments: a suburban, a major highway,
and a city environment. This research shows that LiDD and it's resulting data sets can be
developed without requiring expensive equipment and that it's data will be useful for future
research and development of Advanced Driver-Assistance Systems (ADAS).
