Cognitive Vision Systems
EU-FP5 project IST-2000-29375
Key Area 4: Essential Technologies
About the project
The objective of CogVis project was to provide the methods and techniques that enable construction of vision systems that can perform task oriented categorization and recognition of objects and events in the context of an embodied agent. The functionality will enable construction of mobile agents that can interpret the action of humans and interact with the environment for tasks such as fetch and delivery of objects in a realistic domestic setting.
The project delivered particular basic methods for recognition / categorization of objects and events/actions in large scale scenarios, new methods for robust interpretation of dynamic scenes, methods for acquisition of basic skills and environmental models, and techniques for fully distributed control of continuously operating systems.
The CogVis consortium was composed of the following institutions:
- Royal Institute of Technology, Department of Numerical Analysis and Computer Science, Stockholm, Sweden
- Hamburg University, Computer Science Department, Hamburg, Germany
- Max-Planck Institute for Biological Cybernetics, Dept of Phychophysics, Tuebingen, Germany
- Leeds University, School of Computing, Leeds, UK
- University of Genoa, Dipartimento di Informatica, Sistemistica e Telematica, Genova, Italy
- Swiss Federal Institute of Technology Zurich, Dept of Computer Science, Zurich, Switzerland
- University of Ljubljana, Faculty of Computer and Information Science, Ljubljana, Slovenia
Our group contributed new complementary subspace methods for continuous learning, unsupervised categorisation, and robust recognition.
In particular, we focused on the subspace methods, which proved to be successful both for learning and recognition under constrained conditions. However, their full potential in unconstrained natural environment has not been exploited. Also, traditional subspace methods have not been designed for open-ended learning. Since this is one of the main requirements of a cognitive vision system operating in natural environment, we focused on developing continuous open-ended learning algorithms. Robustness is another main requirement of cognitive vision, therefore we aimed at developing new robust learning and recognition subspace algorithms, which will be able to work in uncontrolled environments. Learning algorithms capable not only of classification but also categorisation of objects and scenes were developed in the framework of the subspace methods.
More specifically, we were developing algorithms for:
- robust subspace recognition
- recognising categories
- continuous learning
- robust learning in unconstrained environments