TOP
Search the Dagstuhl Website
Looking for information on the websites of the individual seminars? - Then please:
Not found what you are looking for? - Some of our services have separate websites, each with its own search option. Please check the following list:
Schloss Dagstuhl - LZI - Logo
Schloss Dagstuhl Services
Seminars
Within this website:
External resources:
  • DOOR (for registering your stay at Dagstuhl)
  • DOSA (for proposing future Dagstuhl Seminars or Dagstuhl Perspectives Workshops)
Publishing
Within this website:
External resources:
dblp
Within this website:
External resources:
  • the dblp Computer Science Bibliography


Dagstuhl Seminar 06311

Sensor Data and Information Fusion in Computer Vision and Medicine

( Jul 30 – Aug 04, 2006 )

(Click in the middle of the image to enlarge)

Permalink
Please use the following short url to reference this page: https://www.dagstuhl.de/06311

Organizers




Press Room

Press Release

Messdaten der Medizin und Umwelt sinnvoll strukturieren Pressemitteilung vom 27.07.2006 (German only)


Summary

1 More Than the Sum of Its Parts

Today many technical systems are equipped with multiple sensors and informa- tion sources, like cameras, ultrasound sensors or web data bases. It is no problem to generate an exorbitantly large amount of data, but it is mostly unsolved how to take advantage of the expectation that the collected data provide more information than the sum of its parts. The design and analysis of algorithms for sensor data and information acquisition and fusion as well as the usage in a differentiated application field was the major focus of the Seminar held in the International Conference and Research Center (IBFI), Schloss Dagstuhl. 24 researchers, practitioners, and application experts from different areas met to summarize the current state-of-the-art technology in data and information fusion, to discuss current research problems in fusion, and to envision future demands of this challenging research field. The considered application scenarios for data and information fusion were in the fields of computer vision and medicine.

2 Applications of Sensor Data and Information Fusion

  • 2.1.Computer Visison
  • 2.2 Medicine

3 Methods

Image registration and data fusion can be considered as an optimization task: a proper objective function is defined, and the the fusion task is solved by the optimization of the objective function. This includes both classification and re- gression problems.

In image registration, for instance, the objective function is defined through distances of assigned image intensities or matched point features. Within the optimization process the transformation parameters are estimated that minimize the objective function. The registration problem based on point correspondences can be considered as a mixed integer optimization problem. Intensity based image registration requires the optimization of an objective function that measures the similarity of intensities of assigned image grid points. Commonly variational approaches are used as well as gradient descent methods. In the seminar it was also shown that the variational formulation can be interpreted in the context of optimal control of partial differential equations. Other contributions have demonstrated that registration can be done in an unified framework with image pre-processing like intensity correction, image enhancement, and segmentation. Prior knowledge for image registration can be generated by hybrid scanners or manually registered data sets. By the incorporation of specific statistically motivated regularization terms, the objective function can take account for priors. The diagnosis of incomplete multimodal image data makes use of priors, too. In the seminar novel statistical learning methods for the analysis incomplete data as well as for the acquisition of priors were discussed.

Statistical approaches contribute to advances in decision making and classifcation in the presence of multiple information sources. The Bayesian theory of designing multiple expert systems provides a formalism for the treatment of sensor data fusion in pattern recognition and opens up new dimensions in classification theory. This technology can be applied to standard pattern recognition problems as well as applications like driver assistance, object tracking or robot attention control.

4 Conclusions

Computer vision problems are traditionally motivated by robotics and surveillance applications. Most medical image analysis problems come from the application fields medicine and biology. Though no less important, the information fusion problems posed by the robotics community tend to be more related to basic research than those arising from medicine. As they are typically dynamic in nature, must work quickly, and must tively deal with rapidly changing and unknown environments. By comparison, the majority of information fusion problems in medicine and biology are static and operate under more control- led conditions. Not surprisingly, techniques to perform information fusion have evolved differently in these communities with minor to no overlap. In an inspiring environment the seminar brought together members of both communities and initialized scinetific discussions that yield hope for the huge potential of synergies.


Participants
  • Ferid Bajramovic (Universität Jena, DE)
  • Gregory Baratoff (Siemens VDO Automotive AG - Lindau, DE)
  • Rainer Benning (Universität Erlangen-Nürnberg, DE)
  • Volker Daum (Universität Erlangen - Nürnberg, DE)
  • Joachim Denzler (Universität Jena, DE) [dblp]
  • Wendelin Feiten (Siemens AG - München, DE)
  • Gernot Fink (TU Dortmund, DE) [dblp]
  • Bernd Fischer (Universität zu Lübeck, DE)
  • Christoph Gütter (Siemens - Princeton, US)
  • Dieter Hahn (Universität Erlangen - Nürnberg, DE)
  • Jingfeng Han (Universität Erlangen - Nürnberg, DE)
  • Jörg Hipp (Universitätsklinikum Hamburg-Eppendorf, DE)
  • Joachim Hornegger (Universität Erlangen - Nürnberg, DE)
  • Olaf Kähler (Universität Jena, DE)
  • Josef Kittler (University of Surrey, GB) [dblp]
  • Günter Leugering (Universität Erlangen-Nürnberg, DE)
  • Björn H. Menze (Universität Heidelberg, DE) [dblp]
  • Jan Modersitzki (Universität zu Lübeck, DE) [dblp]
  • Christoph Munkelt (Fraunhofer Institut - Jena, DE)
  • Gunther Notni (Fraunhofer Institut - Jena, DE)
  • Rudolf Rabenstein (Universität Erlangen-Nürnberg, DE)
  • Daniel Russakoff (Stanford University, US)
  • Michael Stiglmayr (Universität Erlangen-Nürnberg, DE) [dblp]
  • Sven Wachsmuth (Universität Bielefeld, DE)

Classification
  • artificial intelligence / robotics

Keywords
  • sensor and data fusion
  • adaptive fusion
  • multimodal fusion
  • multiple classifier fusion
  • computer vision
  • robotics
  • medical imaging