Institute of
Cognitive Integrated Sensor Systems
Prof. Dr.-Ing. Andreas König
Institute of
Cognitive Integrated Sensor Systems
Prof. Dr.-Ing. Andreas König
Intelligent Sensor Systems
Image Processing/Machine Vision:
Image processing has established itself as an
important technology in industry as well as in research. The field of
measurement and instrumentation more and more commonly employs image
processing to solve measurement tasks. Advances in sensor technology
and integration deliver new sources of pseudo-image data, which in part
are amenable to etablished or modified approaches from image
processing. Recent examples are. e.g., range imaging or emerging
Terahertz-technology. Robot vision is one field, that is strongly
related on the one hand to image sensor and embedded vision system
implementation and on the other hand to image processing. In this
context, ISE has been active for nearly a decade designing
application-specific image processing and recognition systems, e.g.,
for visual industrial quality control, as well as image sequence
processing related tasks, e.g., for driver
assistants or eye-tracker case studies and applications.
The improvement of the design process of such systems is a key topic in
the ISE image processing research. Commonly, such systems are designed
in a tedious, time-consuming, iterative process of try-and-error,
requiring substantial expert knowledge and manpower. The outcome often
was not proportional to the required effort. The development of a
methodology and corresponding tools for systematic and optimized design
of image processing and recognition systems has been pursued by ISE for
more than a 15 years now. The first outcome was the proprietrary QuickCog tool, which stream-lined the design
process, provided numerous novel visualization aids and the first
automation features by employing optimization methods. This has been
employed in numerous industrial and research projects. Currently, also
the lower levels of image preprocessing and feature computation are
subject to automation in doctoral thesis project, employing Genetic
Algorithms and Particle Swarm Optimization. The following selected
projects have significantly contributed to that line of research:
A Reconfigurable, Robust Robot Vision System for Medical Laboratory Automation
Holistic Design of Recognition Systems (Project GAME, SPP 1076 VIVA)
Contributions to the Development of Advanced Knock Detection Methods in Gasoline Engines
DeCaDrive - Multi-Sensor Intelligent System for On-Line Driver State and Intention Recognition
Resource-Efficient Signal Processing & Platforms:
In particular, mobile systems or systems with restricted or no access to the power grid require particular care in design with regard to power consumption and dissipation. Examples of recent fields are Ambient Intelligence (AmI), Wireless-Sensor-Networks (WSN), robotics, or automotive applications, where wired supply and communication would contradict economic and/or constructive constraints. In this research, the design of mobile in particular wireless sensor systems is investigated with regard to a power-efficient or even self-sufficient architecture. Resource-efficient algorithms for signal processing and recognition tasks are one important means. Consequently, the research is linked with activities on sensor system design automation discussed below. Further, distribution of processing on several nodes in a WSN as well as aspects of energy harvesting relate to this research interest. The following projects have so far contributed to that line of research:
A Wireless Color-Sensor Module for Sensor Networks in Ambient Intelligence Applications
A Lego Mindstorm NXT Robot as an Extendable Mobile Sensor Plattform
Holistic Design of Recognition Systems (Project GAME, SPP 1076 VIVA)
Lab-on-Spoon Research Activity in ISE-Smart-Kitchen
E-Taster Assistance System with Lab-on-Spoon and Lab-on Fork as 'Electronic Tongues'
Multivariate Data Visualization & Analysis:
The treatment of high-dimensional, large data sets is a common problem encountered in numerous disciplines, from recognition system design and medical applications to microelectronics manufacturing and circuit design. The data can origin immediately from measurement, be the result of intermediate processing or be meta-data from more abstract levels and processes. The observation of optimization processes can be affiliated to the latter case, whereas features spaces in image or recognition systems affiliate to the second case. Visualization of high-dimensional data requires efficient and meaningful dimensionality reduction methods. Such mapping or projection and the visualization methods and tools are a subject of research at ISE for more than 15 years and is still a major ISE research interest. The QuickCog-System embeds a subset of the most convenient and efficient methods for image and recognition system design. In the following projects a more elaborate set of methods has been developed and applied to interdisciplinary problems:
ABSYNTH: A Methodology and Framework for Transparent and Efficient Analog Circuit Design Automation
Advanced Methods for the Analysis of Semiconductor Manufacturing Process Data
Application of Interactive Multivariate Data Visualisation to the Analysis of Patients Findings in Metabolic Research
Automation of Intelligent Sensor System Design:
The research pursued in this activity is related to the research on image processing and recognition system design, but the focus is on the special needs of multi sensor systems comprised of a potentially heterogeneous collection of sensors. In particular, the influence of sensor properties and scene settings, e.g., sensor heating, on measurement and the spatio-temporal nature of the data impose different constraints and require a distinguished approach. From the feature level to classification, the relation to image processing systems are close and obvious. The same considerations on design effort and efficient procedure apply. However, the nature of intelligent sensor systems and their embodiment in lean integrated (MEMS) or embedded implementation also impose constraints on the optimization in the design process. Resource limitations low-power consumption require the design for a lean yet well performing intelligent integrated sensor system. In particular, gas sensors or eletronic noses are subject of case studies in this research context. The following doctoral project currently contributes to that line of research:
ABSYNTH: A Methodology and Framework for Transparent and Efficient Analog Circuit Design Automation
Advanced Methods for the Automation of Intelligent Sensor System Design
Contributions to the Development of Advanced Knock Detection Methods in Gasoline Engines
Fault-Tolerant, Adaptive, Self-x Sensor Systems:
This recent research activity deals with the design of next generation sensor systems, which incorporate advanced features of error-awareness and -correcting abilities. Sensor as well as sensor electronics suffer from various failure mechanisms and at least partially irreversible performance degradation during system operation. Self-monitoring abilities of a sensor system, e.g., by auxiliar sensors/actuators allow the detection of such errors and deviations. Reconfigurability and redundant resources open the door to self-repair or self-healing. In this research mechnism of sensor system adaptation to environmental changes as well as to sensor & electronics degradation or failures are investigated. The work focuses on issues of importance for integrated and intelligent sensor systems and is closely linked with the activities on dynamically reconfigurable sensor electronics and MEMS design in general. The following projects have contributed so far to that emerging research topic:
A Reconfigurable, Robust Robot Vision System for Medical Laboratory Automation
MEMS-DC-Relays for Dynamically Reconfigurable integrated Mixed-Signal Electronics and Systems with Self-x Properties