Field of Research
Research in the section at the department of computer science focuses on the devising and realization of graphic-interactive systems, in particular the development of graphic algorithms, their efficient conversion to hardware (VLSI) and software, computer architecture for multi-processor solutions, and software concepts to programme them.
Based on this "core graphics", research work is carried out into handling of very large data sets (Multi-Resolution Modeling, Mesh Simplification and Compression), computer animation (physically correct modeling of the dynamics of fabrics), virtual reality (virtual work bench and tactile input), photo-realistic imaging (radio lab), the visualization of scientific data and very complex structures (relativistic effects), minimally invasive surgery and virtual endoscopy as well as hardware for real time volume visualization (VIZARD).
Vision for reconstruction of static and dynamic scenes
Large-scale city and terrain reconstruction aims to provide tools for the interpretation and analysis of images with the goal of generating precise models of the reality depicted. Computer vision is an area where it becomes very clear that human vision in many respects is still far superior to what can be achieved with computer algorithms. One important research question is how to utilize prior knowledge in the process of model generation, and how to acquire such prior knowledge from examples. The detection of general symmetries is a promising approach to improve automatic analysis of geometrical models.
We have successfully applied analysis by synthesis techniques to generate detailed textures from very low-resolution photographs of which each individual shot lacks the details needed. Another area of research in the field of computer vision is the analysis of dynamic scenes using variants of optical flow algorithms. The estimation of motion and ego motion has important applications in the areas of UAV control and driver assistance systems, which are investigated in two PhD projects industry-funded.
Statistical Voxel Models of the Human Body
The automatic interpretation of medical images is another research area where the inclusion of prior knowledge about human anatomy and physiology plays a very important role. We try to improve the processing of medical imaging data using statistical voxel models generated from imaging data from a large number of individuals. Models generated in this way allow the inference of information from sparse measurement data that, although statistical in nature, is valuable for many applications from operation and diagnosis planning via automatic labeling, all the way to diagnosis of certain diseases.
Human Computer Interaction
Human Computer Interaction is an area of computer science that can benefit greatly from results in perception research. On the other hand established techniques, developed in computer science, e.g. techniques of virtual and augmented reality, can be used to systematically investigate the mechanisms of human perception. Several projects aim to achieve progress in both fields: understanding human perception of space in a better way by using virtual reality techniques, and improving the interpretation of user intention with gaze-based interaction techniques, or providing individualized multi-touch interaction.
For the sake of completeness three projects in the context of digital media should be mentioned. They deal with questions of rendering, sound processing and color postproduction of motion pictures, topics that are closely related to human perception. These projects are run within the collaborative PhD programme “Digital Media Production” in collaboration with Hochschule der Medien Stuttgart and the University of Stuttgart.