Location-based services is an upcoming market providing location specific information to the user. While GPS and cellular networks help in rough localization outdoors, they cannot provide directional information and exact position. Cameras do not require any external signals and work almost everywhere. However, the data dimensionality of images makes the data association (matching) along multiple frames hard. We show in this talk that we can solve the visual localization problem without matching with a voting scheme.
Voting of the parameter space (known as Hough transform) can be accelerated if we realize that the transform integral is a convolution integral, not necessarily on the image plane. Using tools from harmonic analysis on groups, we give a new light to Fourier methods and propose algorithms that are robust and suitable when the majority of features are outliers. We close the talk with a new and fully autimatic method for registering to eachother range scans with limited overlap.
Kostas Daniilidis is Associate Professor of Computer and Information Science at the University of Pennsylvania where he was Assistant Professor from 1998 to 2003. He is a member of the interdisciplinary GRASP laboratory. He obtained his MSE (Diploma) in Electrical Engineering from the National Technical University of Athens, 1986, and his PhD (Dr.rer.nat.) in Computer Science from the University of Karlsruhe, 1992, under the supervision of Hans-Hellmut Nagel. His research interests are in space and motion perception with machines, with applications on navigation, omnidirectional vision and immersive environments. He is Associate Editor of IEEE Transactions on Pattern Analysis and Machine Intelligence.
He was the chair of the 2000 IEEE Workshop on Omnidirectional Vision and Area Chair of the ECCV 2004, CVPR 2004, 2005, and 2006 conferences. In June 2006, he co-chairs with Pollefeys the Third Symposium on 3D Data Processing, Visualization, and Transmission.
