Obavijest o predavanju

Abstract: I will start with an overview on previous achievements in active object recognition and in object categorization. It is obvious that existing methods in object categorization are severely constrained by 2D view-based modeling, with only a few very recent exceptions towards 3D category models. In search for a shape-based 3D category model, we are currently investigating various enabling methods. We need to know about camera pose in a scene, and object pose with respect to an observer – a partial solution to this problem is our recent development of a novel online structure and motion algorithm for general camera models. In our current, ongoing work, there are 3 PhD students investigating: (a) 3D shape-based modeling (3D contours and their pair-wise spatial relations), (b) spatiotemporal representation and reasoning, and (c) various ways to interact with a dynamic scene. A combination of these methods will, in the near future, enable us to generate hypotheses about 3D scene content, and to disambiguate/clarify a specific hypothesis by actively taking further views of an object. This novel approach can, on a broader perspective, be viewed as a contribution towards the integration of bottom-up and top-down processes and towards the confluence of “recognition school” and “reconstruction school” in computer vision.

Axel Pinz received the MSc degree in electrical engineering in 1983 and the PhD degree in computer science in 1988 from the Vienna University of Technology. In 1995, he received the habilitation degree in computer science from Graz University of Technology. Since 1999, he has been with the Institute of Electrical Measurement and Measurement Signal Processing (EMT), Graz University of Technology, Austria, where he is heading the vision-based measurement group (vmg), a research group that is focused on real-time measurement and object recognition. His main research interest is in high-level vision including spatiotemporal reasoning and tracking for object categorization and scene recognition. He is a member of the IEEE. Axel Pinz has supervised or co-supervised more than 50 MSc theses and 29 PhD theses students. He has published 28 journal papers and 110 papers at peer-reviewed, international conferences.