Upload
tilman-seubert
View
212
Download
0
Embed Size (px)
Citation preview
DEVELOPMENT Assistance Systems
10 ATZ worldwide 7-8/2002 Volume 104
Image sensors have a good chance of becoming establishedalongside radar and laser-based sensor systems. They facilitatethe optical detection of objects and features in traffic spacewhich can be detected only with great difficulty or not at allusing other physical sensor principles, putting them in aposition to provide effective support for assistance systems ofthe next generation. This article by Hella KG Hueck & Co.explains the sensor characteristics as well as possible futureapplications in some detail.
1 Introduction
The importance and the share of
automotive safety systems on the one hand
and systems providing more comfort and
convenience on the other will significantly
increase in the next few years. Sensors for
measuring data in the vehicle environment
are the first links in the signal processing
chain in all such systems. This results in
two new challenges: the measurement of
physical parameters outside the vehicle
By Martin Mühlenberg
and Tilman Seubert
Fahrerassistenzsysteme
basierend auf Bildsensorik
You will find the figures mentioned in this article in the German issue of ATZ 7-8/2002 beginning on page 658.
Driver AssistanceSystems Based onImage Sensors
11ATZ worldwide 7-8/2002 Volume 104
system and, often inevitably linked with
this, the use of innovative sensor concepts.
With the aid of image-producing sensor
technology and mechanical sight, pictorial
features describing the infrastructure can
be detected in traffic space. Optical sensors
from Hella already used in standard vehicle
models, such as rain or light sensors, form
the optimum basis for integrative and
functional expansion through future image
sensor concepts. Expertise in the field of
optics, synergies within electronic and
software development as well as a
thorough understanding of integrative
questions are essential pre-requisites for
survival in the automotive product market
of future optical-image-producing systems.
Figure 1 illustrates the basic topology of
future driver assistance systems. A
distinction is made between the long-
distance range > 100 m covered by 77 GHz
radar or infrared lidar systems for functions
such as Adaptive Cruise Control (ACC), the
mid-range up to 50m and beyond covered
by image-processing methods for functions
such as Lane Departure Warning or night
vision, and a close-up range covered by 24
GHz radar or infrared scanners for
functions such as stop & go, automatic
emergency brake, pre-crash, recognition of
the blind angle and parking assistance.
Figure 2 classifies the image sensor
applications according to customer
functions, complexity and possible
moment of use: the entry functions are
comparatively easy and cover
conveniences requirements. After this the
driver assistance functions develop into
safety-related networked functions, on the
one hand increasing the complexity of the
algorithms and on the other making the
data fusion of several measuring principles
necessary.
2 Lane Tracking: One of theFirst Applications of ImageSensors in Driver AssistanceSystems
The first image sensor applications in
passenger vehicles are rear-view camera,
night vision and lane tracking. Whereas
with the first two of these image
reproduction for the driver and ergonomic
marginal conditions are at the fore, the last
mentioned feature manages without
image representation. In this last case,
image analysis and the transparent
application for the driver with regard to the
function are at the fore.
Figure 3 illustrates the most important
features of a lane tracking system:
■ The sensor captures the environment in
front of the vehicle. Accuracy and stability
in a wide range of different light and
weather conditions are the most important
features of the image quality. These are
mainly determined by optical and image
sensor chips.
■ Integration in the vehicle. Optimum
fulfilment of the customer function with
aesthetic integration and a structure and
connection technology which is as reliable
as it is economical are the main design
objectives.
■ The algorithms of image processing
mainly characterise the application. These
will be explained in the following in more
detail.
■ The interface to the driver completes
the customer function. It is here that the
largest share of customer-related
differentiation can be found.
3 Sensor Features
For applications with a view forwards, the
obvious place to install such sensors is
behind the wiped area of the windscreen.
On the sensor side, there is a lens unit and
design measures to reduce stray light
situated in front of a CMOS image sensor
chip. Primary design parameters
determined by the application(s) are
mainly the alignment of the optical axis,
the horizontal and vertical aperture angle
and the sensor resolution. The spectral
sensitivity of the image sensor, the high
variability of the spectral distribution of
the light in the environment to be recorded
(sunlight, artificial light at night) as well as
the transmission behaviour of the
windscreen must all be taken into
consideration for the optical design.
Depending on the sensitivity of the sensor
and the aperture chosen, sufficient
exposure must also be ensured at night,
Figure 4 and Figure 5. The type of dipped
beam light source used for illumination of
the scenario (halogen or gas discharge
bulb) has a decisive influence on sensor
exposure due to its different spectral
distribution. Situations at night in
particular make it clear that the
performance of a driver assistance
application based on image sensors is
determined by the limitations with regard
to the prevailing visible conditions. Thus,
for example, the viewing distance for lane
tracking at night without an additional IR
light source is limited to the range achieved
by the dipped beam light. Furthermore,
detection of neighbouring lane markings in
the dark in scenarios with more than one
lane often causes problems, resulting in
limited performance when changing lanes
and putting the search areas for image
processing into practice. On the other hand,
some conditions for feature detection are
better at night, since the retro-reflecting
property of the markings at least supplies
good image contrasts and prevents the
occurrence of shadow situations. One point
often discussed in connection with the
physical radiation properties of sensors for
automotive use is the demand for sufficient
dynamics. Since extreme exposure
situations can often occur extremely
quickly (for example, shadow situations in
a tree-lined avenue at high vehicle speed),
the required high grey value dynamics
must be related to each individual image
frame. Last but not least, the whole camera
unit is subject to great temperature
influences, which result in fading images
in the image sensor chip and material
expansion with corresponding optical
effects in the imaging unit, Figure 6. A
sturdily designed image processing system
can tolerate such effects up to a given limit
depending on the application involved,
since the image has been intended for pure
image processing and does not have to
fulfil the visual requirements of a human
observer.
4 Lane Departure Warning
Unintentional lane departure is one of the
most common causes of fatal accidents. The
reasons for this vary, and include
distraction of the driver and excess speed,
to give only two examples. A system that
detects the position of the vehicle in
relation to the lane markings and compares
this with the expected driver intention,
which can be determined by means of
changes in steering angle, activation of the
indicator light and brake pedal, can provide
additional information to the driver in
critical situations and give a warning. Such
lane departure warning systems are
already available in the commercial vehicle
sector and will enter the passenger car
sector in the next few years. Initially, such
systems will primarily be used in well-
structured scenarios with those road
markings already available. White lane
markings as pictorial patterns allow
reproducible and advantageous
recognition using appropriate image
processing operators. This requirement of
solid feature recognition must not be
underestimated, and guarantees successful
image processing in a wide range of
different lighting and weather conditions.
Edge-based image processing operators
measure the positions of the lane markings
at different viewing distances, Figure 7.
These are entered into a model-based
tracking using a Kalman filter process.
DEVELOPMENT Assistance Systems
12 ATZ worldwide 7-8/2002 Volume 104
Descriptive model parameters are the lane
geometry and the vehicle kinematics.
Special features of the infrastructure in the
course of the lane such as filter lanes and
motorway exits often require a regulation-
based feature selection as an initial
parameter component, due to the
ambiguity of the markings, Figure 8. The
requirements with regard to aperture angle
are first of all dependent on the application.
In the case of lane tracking, this would be
the detection of markings as near to the
vehicle as possible – thus describing the
position of the vehicle itself very well –
resulting in a wide-angle design of the lens.
Manufacturing and installation tolerances
as well as vehicle-dynamic structural
movements must also be taken into
account, however, in that a certain increase
in angle must occur with regard to the
design of horizontal and vertical aperture
angle.
5 Object Recognition
In contrast to the individual sensors with
their primarily specific applications, the
driver assistance systems of the future will
require an overall view. Thus, future image
sensor systems will initially detect the lane
and warn drivers if a lane is changed
unintentionally and will then in future
sensor generations certainly take over
further tasks in combination with other
sensor systems. One of these applications
will be supportive object recognition for
ACC systems, which can already be found
in the form of radar or laser sensors in some
vehicle models today. The distance and
lateral position of vehicles and objects
supplied by the ACC sensor serves as a basis
for fixing parameters for good pre-
conditioned image processing with regard
to the values to be expected. This is then
used as a basis for recognising object
contours through the image sensor system.
In this way, the search areas in images and
contour positions to be expected can be
restricted in a well-defined way. The results
are more exact lateral position and
dimension data of the objects followed by
the ACC, as well as a more accurate
estimation of the cut-in and cut-out
manoeuvres of the relevant vehicle targets.
Vehicles can be classified with regard to
their design in order to implement a model-
adaptive strategy with length regulation
and, last but not least, when coupled with
the lane tracking function, the system can
allocate lanes to the relevant vehicle
targets.
6 Further Functionalities
The present Hella light sensor uses two
independent receiver elements for
luminance detection in the environment
and directly in front of the vehicle. The
locationally resolved measurement
achieved using an image sensor optimises
the detection of tunnels and underpasses in
particular. Furthermore, such a sensor
plays an important role in the selection of
situation-adapted beam patterns. Absolute
measurements are possible, taking sensor
exposure parameters and appropriate
calibration into account. The knowledge
gained about the environment by an
image-processing-based light sensor
system leads once again to even better
settings of the parameters for the image
processing algorithms for the lane tracking
application. This mutual support and the
similarity of the required optical marginal
conditions says a lot for combining these
two applications in one image sensor
system. This is why the functional
possibilities and sensible focusing of
different applications into one image
sensor system are being pursued in the
development steps.
7 Block Structure of an Image Sensor System for Lane Tracking
Figure 9 represents the basic structure of
an image sensor system:
■ The front-end consists of a lens and an
image sensor. The image captured is / not
pre-processed and forwarded to the
computer unit via a suitable image data
bus.
■ Image processing and application
algorithms run on a digital signal processor
(DSP).
■ Communication with the vehicle
interface to the HMI can take place via CAN.
Depending on the construction space
available, the image processing system can
be partly or completely integrated in a
different module. Thus, for example,
solutions are currently under consideration
in which the image sensor front-end is
integrated in the roof control module, as
illustrated in Figure 2, whereas the
computer unit can be installed elsewhere.
In this case the image data bus is subject to
higher EMC requirements.
8 Summary and Outlook
The reliability of lane tracking has been
proved using extensive driving situations.
Based on these results, Hella is working at
high priority on concrete implementation
in an automotive product. The introduction
of a series product is to be expected in 2005.
Lane tracking is certain to be one of the first
applications. Once image sensors have
been generally accepted in vehicles, further
applications will follow. The triggering of
vehicle lighting is important in this context
in particular. The future networking of
environment sensors will then be the next
step towards a complete driver assistance
strategy. Due to the complementary sensor
characteristics, the combination of an ACC
sensor and an image sensor system creates
good pre-conditions for optimised vehicle
and object recognition, Figure 10. The aim
is to record the environment of a vehicle in
its whole complexity in order to achieve
functions which initially increase
convenience and subsequently safety. ■
DEVELOPMENT Assistance Systems