Collision
threat detection and avoidance defines a major Research and
Development challenge for the automotive industry. Adaptive
cruise-control systems incorporating some collision-avoidance
features are offered today as pricey options on luxury cars. However,
the performance of these systems is not always sufficient, and their
cost is too high for wide use. Significant improvements are still
needed for these systems to perform satisfactory and to become
popular.
Future
cruise-control systems will probably fuse data from different types
of sensors, with optical images, and therefore vision,
playing a significant role. However, present
conventional
approaches to vision, consisting of a camera that acquires the data
and a separate digital processor that process it, are too slow for
the most demanding tasks. New solutions are needed. This project
brings together a multi-disciplinary team including
mixed-signal microelectronic designers (IMSE-CNM), one company in the
automotive sector (VCC), neurobiologists (UNEW), and experts on
opto-electronic information technology and hardware/software systems
(ANCL-HAS) to target the development of these new solutions.
The
project embraces advanced research and development activities focused
on creating single-chip bio-inspired visual perception systems for
automotive applications. These systems will be based on the
outstanding performance of natural visual sensory-processing systems.
As a central reference point, the project focuses on the integrated
visual neuro-system for collision avoidance found in grasshoppers,
which will be studied, modelled and emulated by means of
sensory-processing chips in standard CMOS technologies. The resulting
electronic systems will be able to operate correctly within the wide
range of environmental conditions encountered in real-life automotive
applications, and will meet the strict reliability standards of the
car industry.
Thus,
this Project’s global objective is the conception and
development of architectures and general design techniques for
bio-inspired Vision Systems on Chip (VSoC) integrating
sensing, perception and action-control features. In order to
demonstrate the reliability of the developed techniques under real
conditions, and as a way of measuring the project achievements, a
demonstrator intended for course-to-collision signalling and
avoidance in automobiles will be designed, fabricated, tested and
evaluated under real world operating conditions. In this sense, the
successful achievement of this demonstrator can be considered as a
central project objective. However, the fundamental objectives are
far more ambitious, since VSoCs designed with the same basic
techniques to be developed will find applications in many other
tasks, within the automotive field and in many other fields. In this
other sense, the success of the project should be measured against
the validity of the devised architectures and circuit techniques for
similar applications.
The
aimed VSoCs differ from conventional image sensors in that they will
merge the following concurrent functions on a common silicon
substrate:
-
Sensory
circuitry;
-
Processing
circuitry for, on the one hand, adapting the sensors response to the
changing environment and; on the other hand, analysing the
spatial-temporal characteristics of the visual flow, extracting the
relevant information, and transmitting it to a higher-level processor.
-
High-level
processor and actuator-control for action-oriented decision-making and
driving.
The
partners are confident that the achievement of the project objectives
will lead to significant advances in automobile active-security and
other technically-similar applications in many diverse fields. The
reasons behind this confidence are related to performance and cost.
Present aproaches to the use of vision in active car security employ
conventional digital cameras and subsequent digital processing of the
frame sequence. This approach is generaly slow, bulky, and expensive.
For these reasons, this type of device has not made its way into
widely available commercial products, despite the increasing value of
safety in the automobile industry. The use of specific VSoCs will
reduce the cost and physical sizes of the devices, and will also
improve performance exploiting the new capabilities of CMOS
technologies for image sensing and for concurrent processing.
On
a lower and more specific level, the project global objective can be
decomposed into the following sub-objectives,
-
At the biological level: extraction, comprehension, and
modelling of the functional aspects of the locust looming-alert
biological system, with emphasis on the underlying mechanisms
of robustness. Analysis of similar biological models for other
functions in other insects, generalisation, and guidelines for
versatility and robustness of operation.
-
At the information technology level: system
characterization of the underlying biological model, embodiment
for electronic implementation, identification of
relevant parameters, and VSoC software and prototyping/testing hardware
development.
-
At
the chip design level: identification of an optimum set of functional
primitives and corresponding circuit implementations, with emphasis on
adaptive sensing, optimal sensory/processing integration, and a smart
synergy of analogue and digital techniques.
-
At
the application level: VSoC specification, including operating-ranges
and conditions, and system integration and evaluation under real
conditions.
Each
of these sub-objectives defines a technical work package, led by a
corresponding partner. Thus, activities related to the modelling of
the vision system of the insect Locust (Locusta
Migratoria) will be basically realized by neuro-biologists
(UNEW); architectural issues will be the responsibility of
information technologists (ANCL-HAS); the design of the custom chips
following this architecture will be done by mixed-signal chip
designers (IMSE-CNM); and finally, car builders (VCC) will be in
charge of defining the system-level specifications and of evaluating
the demonstrator. Because VSoCs constitute the nuclear component of
the proposal, IMSE-CNM has been selected to coordinate the project.
|