Visual Cognitive Systems Laboratory

Visual Cognitive Systems Laboratory (ViCoS) was officially founded in November 2004. Most of the research activities described in this Survey were performed while the members of ViCoS were still with the Computer Vision Laboratory.

Visually enabled cognitive systems are intelligent systems that use vision among other sensors in order to act and interact in everyday situations that emerge in natural and urban environments. This includes a plethora of devices, ranging from mobile robots to intelligent environments, personal devices, and cognitive assistants. The Visual Cognitive Systems Laboratory is involved in basic research in such systems, with emphasis on visual learning and recognition. Other activities include panoramic imaging for mobile robotics and range image modeling and interpretation.

Research in the area of visually enabled cognitive systems focuses on various theories, at different levels of abstraction, regarding requirements, architectures, forms of representation, kinds of ontologies and knowledge, and varieties of mechanisms relevant to integration and control of vision systems. In this context, cognitive vision implies functionalities for knowledge representation, learning, reasoning about events and structures, recognition and categorization, and goal specification, all of which are concerned with the semantics of the relationship between the visual agent and its environment. This requires a vast effort in a multidisciplinary understanding of cognitive processes, involving studies in cognitive psychology, neuroscience, and philosophy of mind.

Specifically, research in the area of visual learning and recognition has focused on subspace methods, such as Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Independent Component Analysis (ICA), Canonical Correlation Analysis (CCA), Support Vector Machines (SVM), etc., which enable direct view-based building of visual representations and subsequent visual recognition of objects, scenes, and activities. Our main research achievement in the framework of subspace methods is development of robust approaches to both learning and recognition. Applications include recognition of objects, scenes, and activities in visual cognitive tasks, such as surveillance and smart vision-based positioning using wearable computing in urban environments as well as in other applications of cognitive systems, such as mobile robots and cognitive assistants.

Our theoretical findings on visual learning and recognition very often ground in a realistic scenario of spatial orientation of mobile robots, which represent a target platform for many of the methods developed. In long terms, we aim at developing algorithms for autonomous exploration and building of topological maps which can be used in cognitive agents for autonomous navigation in unbounded environments. Such cognitive agents will ultimately be able to perceive and understand their environment, to categorise and recognise objects and subjects around them as well as actions they are performing, and will be able to interact with the environment and communicate with humans in a user friendly way.

Research in the area of range image interpretation includes range image acquisition, segmentation of range images using the “recover-and-select” paradigm and modeling of shapes using different types of parametric models. Possible applications include automatic creation of CAD models for reverse engineering applications, creation of models for virtual reality applications, and part-based object recognition.


Selected References

  • A. Leonardis, A. Gupta, and R. Bajcsy. Segmentation of range images as the search for geometric parametric models. International Journal of Computer Vision, 14: 253–277, 1995.
  • R. Bajcsy, S. Wook Lee, and A. Leonardis. Detection of diffuse and specular interface reflections and inter-reflections by color image segmentation. International Journal of Computer Vision, 17(3): 241–272, 1996.
  • A. Leonardis, A. Jaklič, and F. Solina. Superquadrics for segmentation and modeling range data. IEEE Transactions on Pattern Recognition and Machine Intelligence, 19(11): 1289–1295, November 1997.
  • F. Solina and A. Leonardis. Proper scale for modeling visual data. Image and Vision Computing Journal, 16: 89–98, 1998.
  • H. Bischof and A. Leonardis. Finding optimal neural networks for land use classification. IEEE Transactions on Geoscience and Remote Sensing, 36(1): 337–341, January 1998.
  • A. Leonardis and H. Bischof. An efficient MDL-based construction of RBF networks. Neural Networks, 11(5): 963–973, July 1998.
  • W.G. Kropatsch, A. Leonardis, and H. Bischof. Hierarchical, adaptive, and robust methods for image understanding. Surveys on Mathematics for Industry, 9: 1–47, 1999. Springer-Verlag.
  • H. Bischof, A. Leonardis and Alexander Selb. MDL principle for robust vector quantisation. Pattern analysis and applications, 2(1): 59–72, 1999.
  • A. Leonardis and H. Bischof. Robust recognition using eigenimages. Computer Vision and Image Understanding, 78(1): 99–118, 2000.
  • A. Jaklič, A. Leonardis and F. Solina. Segmentation and Recovery of Superquadrics, volume 20 of Computational imaging and vision. Kluwer, Dordrecth, 2000.
  • A. Leonardis, H. Bischof, and J. Maver. “Multiple Eigenspaces”, Pattern Recognition, 35, no. 11, pp. 2613–2627, 2002. Twenty-Ninth Annual Pattern Recognition Society Award. Selected as the most original manuscript from all 2002 Pattern Recognition issues.
  • T. Werner, T. Pajdla, V. Hlavác, A. Leonardis, M. Matoušek. Selection of reference images for image-based scene representations. Computing, vol. 68, pp. 163–180, 2002.
  • J. Maver, A. Leonardis. Recognizing 2-tone images in grey-level parametric eigenspaces, Pattern recogition. letters, 23, pp. 1631–1640, 2002.
  • M. Jogan, E. Žagar, A. Leonardis. Karhunen-Loeve expansion of a set of rotated templates. IEEE Transactions on image processing, 2003, Vol 12, No 7, pp. 817– 825, 2003.
  • M. Jogan, A. Leonardis. Robust Localization using an Omnidirectional Appearance- based Subspace Model of Environment. Robotics and Autonomous Systems, Volume 45, Issue 1, pp. 51–72, Elsevier Science, 2003.
  • B. Kverh, A. Leonardis. A generalisation of model selection criteria. Pattern analysis and applications, pp. 51–65, 2004.
  • H. Bischof, H. Wildenauer, A. Leonardis. Illumination insensitive recognition using eigenspaces. ComputerVision and Image Understanding, Volume 95, no. 1, pp. 86–104, 2004.
  • D. Skočaj, H. Bischof, A. Leonardis. A robust PCA algorithm for building representations from panoramic images. Computer vision - ECCV 2002: proceedings: part IV, (Lecture notes in computer science, 2353), pp. 761–775, 2002.
  • D. Skočaj, A. Leonardis. Weighted and robust incremental method for subspace learning. Ninth IEEE international conference on computer vision, 13–16 October 2003, Nice, France. Vol. 2., pp. 1494–1501, 2003.
  • M. Peternel, A. Leonardis. Visual learning and recognition of a probabilistic spatio-temporal model of cyclic human locomotion. ICPR 2004, pp. 146–149, 2004.

How to reach us?

The Visual Cognitive Systems Laboratory is located on the second floor of the new FRI building on Večna pot 113, Ljubljana.


Current Projects

Closed Projects

Web Page

More about this laboratory