Special Combined VASC/Robot-Learning Seminar ====================================== Date: Friday Jan 21 2000 Time: 3:30-4:30 Place: Adamson Wing (Baker 136a) Speaker: Joachim Denzler (University of Rochester/Universitt Erlangen-Nrnberg) (http://www5.informatik.uni-erlangen.de/HTML/English/Persons/MA/de/de.html) Title: Combining Computer Graphics and Computer Vision: Visual Self-Localization Using Particle Filters Abstract: One important aspect of autonomous mobile system is the so called self localization, i.e. the computation, estimation and tracking of its own position. In this talk a combination of computer vision and computer graphics is presented for reliable and robust estimation based on visual information. The so called lightfield approach for image based modeling and rendering is used to reconstruct an environmental map. This map is then taken as the underlying model for computing photorealistic virtual views of the scene. These virtual views and the actual images taken from the scene are used as a likelihood measure for a probabilistic position estimation method based on particle filters. Due to the modeling technique a very fast rendering mechanism is provided, which can be done in real--time using 3--d texture hardware. Experimental results show the advantages and problems of such an approach.