|
SAVE-U's
challenging and innovative approach is significantly advanced with respect
to the state of the art.
The
project is specially designed for solving at least most of the presently
existing shortcomings.
On
the one hand, SAVE-U will benefit from the other projects related to sensor-based
detection, which holds for a highly efficient development of the entire
sensing platform. On the other hand, SAVE-U fills exactly the gap between
the research projects dealing with the sensor based detection, in particular,
it will provide excellent detection properties: an extremely high detection
rate and at the same time an absolutely low false alarm rate.
In
the following, the 11 items of innovation addressed in SAVE-U are described
in detail.
1. Integrated approach
for protecting the unprotected traffic participants
2. High performance sensing platform
optimised for the detection of unprotected road users
3. Development of a radar network
composed of several 24 GHz sensors
4. Development of 24 GHz sensors
specially designed for the detection of unprotected road users
5. Development of an innovative imaging
system composed of novel IR and vision systems
6. Definition of a new IR camera
system specially designed for the detection of vulnerable road users
7. Development of signal processing
algorithms for the IR subsystem
8. Development of a dedicated real-time
embedded image processing platform for image segmentation
9. Development and implementation
of software algorithms for the classification of unprotected road users
10. Collection of large databases
with ground truth for offline quantitative performance analysis and algorithm
optimisation
11. Development of validation test
procedures and associated test equipment.
1.
Integrated approach for protecting the unprotected traffic participants
Compared to the state of the art, SAVE-U is the first project providing
an integrated approach for the protection of vulnerable road users.
Besides the development of the sensor platform, a warning and actuator
safety concept will be developed and implemented on the demonstrators.
The integrated SAVE-U system concept will assume 4 zones:
- reliable detection for vulnerable road users
- warning the driver
- collision avoidance measures typ. few meters before crash, e.g. emergency
braking
- definition of protection strategies in the case the crash cannot be
avoided (e.g. windscreen or front bumper airbags or adjustable motor hoods.
 |
2.
High performance sensing platform optimised for the detection of unprotected
road users
Safety systems require a highly reliable sensor platform.
SAVE-U utilises a novel approach to reach the goal of reliability: Sensor
fusion both at low and at high level.
High level data fusion, meaning merging of object lists from different
sensors was attempted in earlier research projects. In spite of some improvements
in terms of performance, high level data fusion alone is not sufficient
to provide the required quality and reliability of the target data.
SAVE-U will therefore introduce a novel concept of low level data fusion.
Sensor raw data will be exchanged between the image processing part and
the radar processing part of the sensor platform. Exchange of information
at low level will help a lot to improve the quality (in particular the
detection rate versus false alarm rate) of the objects detected by sensors.
Compared to other projects, this proposal utilises a completely different
sensor system: SAVE-U operates 3 physically different technologies in
parallel and fuses their data: an uncooled IR camera, a network of 24
GHz radar sensors, as well as a video based camera system including improved
signal processing. The SAVE-U platform shall be capable of providing robust
information in all weather and in all lighting conditions.
|
3.
Development of a radar network composed of several 24 GHz sensors
In contrast to the state of the art in terms of 24 GHz radar sensors,
in this project several 24 GHz radar sensors will be used in parallel.
These sensors will have largely overlapping detection areas. Multi-sensor
processing algorithms (advanced versions of triangulation) will be developed
in order to operate the individual single beam sensors in the radar network,
the radar network will provide information about the angle of an object
relative to the vehicle, which is important for the data fusion algorithms.
A dedicated interface to the image processing part of the sensor systems
for low level data fusion will be implemented.
|
4.
Development of 24 GHz sensors specially designed for the detection of
unprotected road users
Compared with the state of the art and with other existing projects, SAVE-U
requires 24 GHz radar sensors with significantly improved range and sensitivity.
Dedicated algorithms for the detection of unprotected road users have
to be developed and implemented. In addition, the update rate has to be
improved. The required, advanced sensors need a complete redesign of the
existing 24 GHz sensors.
|
5.
Development of an innovative imaging system composed of novel IR and vision
systems
Up to now, no project is known that utilises passive IR and video cameras
at the same time, which represents one major advance in SAVE-U. While
passive IR is ideally suited for bad weather and bad lighting conditions,
video cameras provide excellent resolution in lateral and vertical directions.
Combination of both technologies holds for reliable and precise detection
in all weather conditions. Furthermore, it offers the possibility of developing
dedicated image segmentation algorithms that are based on merging the
information from both camera systems and are optimised towards the detection
of unprotected road users.
|
6.
Definition of a new IR camera system specially designed for the detection
of vulnerable road users
Compared to the state of the art (in particular based on the results of
DARWIN and ICAR), in this project the structure of a new IR camera specially
designed for detection of pedestrians and cyclists will be developed.
Due to the high costs for the realisation of the new sensor hardware,
SAVE-U decided to carry out developments only up to simulation level.
Simulations based on real data acquired utilising existing IR camera systems
will be able to clearly show, what kind of images the newly designed IR
system will provide at the end.
|
7.
Development of signal processing algorithms for the IR subsystem
For the first time, the existing image segmentation algorithms for visible
camera systems will be adapted to the images of a passive IR camera. Another
novelty in this field is resulting from merging the IR segmentation results
with those delivered by video cameras and implementing these algorithms
on a dedicated real-time embedded image processing (EIP) hardware.
8.
Development of a dedicated real-time embedded image processing platform
for image segmentation
Compared to the state of the art, in SAVE-U real-time operation is considered
to be very important. To achieve this goal, an embedded image processing
platform (EIP) will be developed and realised. The EIP will be capable
of computing all the image detection algorithms in real-time. Prototypes
of the EIP platform including all the implemented segmentation algorithms
will be integrated into the SAVE-U experimental cars.
|
9.
Development and implementation of software algorithms for the classification
of unprotected road users
A major advance compared to the state of the art will be achieved by the
development and realisation of dedicated algorithms for the classification
of vulnerable traffic participants, such as pedestrians or cyclists. In
particular, reliability of the classification will be significantly improved
compared to the state of the art. Basis is the consideration of the detection
results of both IR and visible light camera subsystems.
The state of the art is represented by algorithms for the detection of
unprotected traffic participants based on single cues (depth, motion,
shape and texture).
Algorithm efficiency and detection performance will be enhanced by:
(a) hierarchical and probabilistic approaches to stereo, optical flow
and shape matching
(b) component-based approaches which are robust to the partial occlusion
of pedestrians and other unprotected road users (e.g. pedestrian behind
a parked car)
(c) identification of the most appropriate pattern classifier for pedestrian
task (e.g. Support Vector Machines, Neural Networks, Radial Basis Functions,
Polynom Classifiers) and associated data dimensionality reduction techniques
(e.g. PCA, ICA).
In addition, multi-cue detection algorithms, significantly improving ROC
performance (correct detection versus false positives) will be developed.
Closed loop tracking strategies will be implemented in order to improve
detection performance by means of prediction.
|
10.
Collection of large databases with ground truth for offline quantitative
performance analysis and algorithm optimisation
A large database of pedestrian data will be collected, that - for the
first time - allows comparative and quantitative analysis of sensor performance.
SAVE-U will make this large database available on the web at the end of
the project. This will allow others to stress-test their pattern recognition
algorithms on a real problem. Utilising EU wide collaboration over the
web, innovative solutions can be introduced in accelerated fashion.
|
11.
Development of validation test procedures and associated test equipment
SAVE-U seeks to demonstrate the improved sensing system by also developing
an innovative validation test procedure. This shall challenge the SAVE-U
system and other systems to demonstrate acceptable performance under true,
real world conditions. The methodologies and technologies required shall
be unique in this field in that they shall allow the validation of fused
sensor systems as well as individual systems. The fundamental methodology
is also flexible enough to allow changes and future improvements as new
issues and sensor technologies become clear.
|
