The 10th European Conference
on Computer Vision
October 12-18, 2008, Marseille, France

ECCV 2008 Workshop on
'Vision in Action: Efficient strategies for cognitive agents in complex environments'


Objectives:

The field of computer vision has developed significantly over the past decades and its importance in fields of development of cognitive systems in the areas of robotics and artificial intelligence is growing. Due to many practical applications in a diverse variety of sub-disciplines, such as surveillance, robot control, virtual environments and medical applications, vision algorithms have a significant application impact. The robotics and computer vision communities have, however, remained largely separate. Dialog between the two communities would contribute greatly to the progress in both disciplines and also open for a fruitful collaboration in terms of integrated projects supported by the European Commission. Today, it is unfortunately a fact that the role of computer vision is not taken seriously into account and that, in the case of cognitive systems aspects, it is seen as 'yet another sensor'. The first goal of this workshop is to initiate a close dialog between the communities and allow for longterm planning of synergistic effects from the integrated research.

In addition, in many cases, methods available in the computer vision community are not in or of general use in the robotics community. On the other hand, the system level perspective is often neglected in the computer vision community, because the research is focused on individual problems and algorithms. Despite having common interests, however, "pure" computer vision has seen significant theoretical and methodological advances during the last decade which many of the robotics researchers are not fully aware of. On the other hand, the manipulation and control capabilities of robots as well as the range of application areas have developed greatly. In robotics, vision can not be considered an isolated component, but it is instead a part of a system resulting in an action. Thus, in robotics the vision research should include consideration of the control of the system, in other words, the entire perception-action loop. A holistic system approach would then be useful and could provide significant advances in this application domain. There is consequently a need to bring the communities of robot and computer vision to a joint appreciation of the value of systems, where there is a need to consider all aspects from perception to action generation.

As a continuation of the ECCV workshop series on Cognitive Vision (2nd Cognitive Vision Workshop), this workshop aims to bring together researchers from various programs and projects working on novel approaches for cognition in technical systems. The second objective of the workshop is to discuss ways of how to integrate biological and psychological findings from human and insect perception and action generation, their exploration and representation strategies. Recent results in combining biological insights into development of perception and action algorithms promise a large impact on research in several fields. Future algorithms need to work in complex environments with limited processing and sensing capabilities that require intensive studies, more specifically related to:

  • Attention control - the complexity of the scene requires additional image processing for a light process decision about areas of interest in the sensor data. Analogies from biological systems presented in the recent publications promise here a significant reduction in requirements on the processing power;
  • Collaborative sensing with simple agents - small, light-weight systems do not allow an implementation of large and complex sensing systems. Collaboration between simple systems is a promising way to tackle this problem;
  • Novel sensor systems - this includes not only a development of new sensors but often a definition of appropriate sensor models from existing physical sensing modalities;
  • Multi-modal fusion of sensory perception - we aim to invite researchers working on fusion of different sensor data representations including physical sensing, resolution, accuracy;
  • Modeling and learning of perception and action for development of innate behaviors - an example of the emerging research in this area are the established vision-based control approaches in the robotics. Recent research results incorporate biological insights in this field re-projecting the data into a simpler image domain or spatial subspace to emphasize signal components in the sensor data that allow a simple definition of action;
  • Memory and representations - coupling between visual perception, tasks, knowledge and the visual system requires memory and suitable data representations. Issues that are of special importance for integrating memory into vision systems include: how to manage representations in presence of limited resources; model for attention; integration of information across representations and time.

Cognitive vision is a research area that combines interdisciplinary research. We hope that the workshop will continue the cross-fertilization between different research communities, research projects, and research groups in the area of computer vision, robotics, artificial intelligence and neuroscience.