2013 Academic Year Seminars
In this talk I will give an overview of my recent research activities at the Center for Vision, Speech and Signal processing at the University of Surrey, UK. I will then talk in more detail about our project to develop recognition software for autonomous reconnaissance robots. The project consisted of development of a swarm of autonomous flying vehicles navigated via wireless link and a GPS system. The UAVs acquire images which are then processed in the ground station by threat recognition software. Our recognition system is based on a vocabulary forest of local features. Large numbers of features are extracted from image examples and are represented by many vocabulary trees. Features from a query image are matched to the trees and vote for object categories and their locations. Large number of trees make the process efficient and robust. The system is capable of simultaneous categorization and localization of objects. The approach obtains excellent performance on large test sets. We demonstrate the robustness of our method to appearance variations,scale change, asymmetric objects, background clutter and occlusion. The system has been extended to human action recognition by incorporating motion information into local features. The performance has been evaluated on 17 challenging real action categories from Olympic Games.
Krystian Mikolajczyk is a Senior Lecturer in Robot Vision at the Center for Vision, Speech and Signal processing at the University of Surrey, UK. He did his PhD at INRIA Grenoble on invariant interest points and then held post-doc positions at INRIA,University of Oxford and Technical University of Darmstadt, Germany,working primarily of image recognition. He currently leads a group of PhD students and research assistants focused on visual recognition problems including issues like image categorization,object and action recognition, retrieval, matching as well as image enhancement, kernel based classifiers, online learning etc.