Two invited talks will be given at the Re-Id 2012:
Rama Chellappa, Center for Automation Research, University of Maryland, USA
Re-identification refers to the problem of recognizing a person at a different location after one has been captured by a camera at an early location. Domain adaptation is an emerging topic in computer vision which tackles the problem where data in the target domain (different location) are not drawn from the same distribution as the source domain (early location). In this talk, we formulate the re-identification problem as an unsupervised domain adaptation problem, where no labeled data are available in the target domain. We discuss several methods for learning a set of intermediate domains between the source and target domains through Grassmannian representation and optimization and sparse coding, and then generate data associated with those intermediate domains. The intermediate data thus generated is used to build a classifier for re-identification of faces.
Anton van den Helgen, University of Adelaide, Australia
Large-scale surveillance offers particular challenges for target re-identification, not least of which are the large number of cameras and targets involved. Airports, for instance, often exhibit tens of thousands of cameras and hundreds of thousands of targets. The limitations of appearance-based measures in such situations need to be understood, and mitigated. We argue that an understanding of how potential targets and cameras relate physically is essential to achieving practically useable results, as is a level of interaction. We demonstrate our findings with respect to a real installation and trail of software implementing the proposed approach within a major international airport.