09-02-2013, 12:49 PM
Detection of Driver Unawareness Based on Long- and Short-term Analysis of Driver Lane Keeping
Detection of Driver .pdf (Size: 93.28 KB / Downloads: 25)
Abstract
Many traffic accidents are caused by driver unawareness. This includes fatigue,
drowsiness and distraction. In this thesis two systems are described
that could be used to decrease the number of accidents. In the first part of this
thesis a system using long-term information to warn drivers suffering from
fatigue is developed. Three different versions with different criteria are evaluated.
The systems are shown to handle more then 60% of the cases correctly.
The second part of this thesis examines the possibilities of developing a
warning system based on the predicted time-to-lane crossing, TLC. A basic
TLC model is implemented and evaluated. For short time periods before lane
crossing this may offer adequate accuracy. However the accuracy is not good
enough for the model to be used in a TLC based warning system to warn the
driver of imminent lane departure.
Introduction
Every year thousands of people die in traffic accidents caused by unintended
lane departure. Statistics from ADAC [11] show that roughly 66 000 or 15%
of single vehicle accidents in Germany occur because of unintended lane departure.
In the USA the numbers are as high as 24% [5]. The reasons for
the driver to unintentionally depart the lane can be many, for example activities
such as eating and drinking but also physical reasons such as drowsiness
and fatigue. These numbers would definitley be smaller if there was a system
in vehicles that warn the driver before the lane departure occurs. In this
master thesis two different warning systems for driver unawereness are examined
and evaluated. The first system evaluates the drivers lane keeping over a
longer time period in the past to detect driver fatigue. The second system uses
current lane keeping information to calculate a Time-to-Lane Crossing, TLC,
value to predict future lane departure.
Objective
The objective of this thesis is to develop two systems to detect driver unawareness
based on driver lane keeping. The first system shall use long-term
information of the drivers lane keeping to warn drivers suffering from fatigue.
The second system shall use short-term information and TLC to predict lane
departure and warn the driver.
Method
The systems are developed with a possible implementation in a vehicle in
mind, which means that calculations and memory usage must be kept at a
minimum. Calculations that would demand large amounts of data buffering
are approximated with incremental functions. Implementation in a vehicle
also means that the calculation parts of the models have to run in real-time. To
allow for this during development the systems are simulated offline in Matlab
Simulink. All blocks are implemented discrete since only sampled signals are
used. All signals are seen as constant between samples. As equation solver a
discrete fixed step method is used.
LAA - Lane-based Attention
Assistant
A lane-based attention assistant, LAA, system to detect driver fatigue is developed.
The system uses three different methods to detect driver fatigue.
One method based on vehicle closeness to lane edge, one method based on
time-to-lane crossing, TLC, and one method based on vehicle over run area,
ORA. In the development of the LAA system the two first methods showed
more potential then the calculation of ORA. As ORA was developed within
another thesis [4] parallel to this one it was decided to concentrate efforts to
closeness to lane edge and TLC. For the sake of completeness the calculation
of ORA is also included.
To filter the input signals for the LAA systems a filter is also developed.
The filter generates a system activity signal that controls LAA system activity
and allows the system to be active only when input signals are usable.
Closeness to Lane Edge
A good indication of sleepy driving is lane drifting [8]. One way of detecting
this is to calculate the distance between the vehicles outer edge and the lane
edge. The minimum distance to lane edge, for the right and left side of the
vehicle, is then delivered every ten seconds. These distances are used to describe
the drivers lane keeping. To decide whether the distance given is close
to the lane edge or not virtual lane edge zones, VEZs, are established. In the
VEZs a weight function is used to grade the vehicles closeness to lane edge.
These weights are then summed and compared to a warning level in order to
decide whether to warn the driver or not. The warning level is individualized
to allow for drivers different driving behavior.
Baseline
By comparing the weight sum, wsumt, from equation (2.11) with a warning
level the system decides whether to give the driver a warning or not. The
warning level is generated individually for every driver and is made up of
a baseline value and a baseline factor. The first part of every test drive is
used to establish a baseline for the driver, i.e a value to describe the drivers
normal closeness to lane edge when not suffering from fatigue.
TLC
First introduced by Godthelp et.al. (1984) (as cited in [13]) the Time-To-Lane
Crossing, TLC, value represents the time available for the driver before any
part of the vehicle crosses a lane edge. Throughout the years different models
for the calculation of TLC have been suggested. The most basic model in [13]
defines TLC as the lateral distance to lane edge divided by the vehicles lateral
velocity. This is also the model used here. A closer description of TLC and
the exact calculation of the TLC value can be found in sections 6.1 and 6.2.