Towards an Intelligent Cognitive Vision Platform

 

Monique Thonnat and Francois Bremond

 

1 Definition

 

An intelligent cognitive vision platform is a scene understanding system with a full degree of autonomy to adapt itself to new situations, with high interaction capabilities using a friendly language to understand user needs and  with various ways to communicate its understanding of the observed scenes.

 

2 Detailed functionalities

 

2.1 Autonomy

The intelligent cognitive vision platform will have autonomy capabilities in two different modes: static configuration or dynamic reconfiguration. In a configuration phase the platform can select automatically pertinent vision algorithms from a library of programmes and can process data coming from a set of heteregenous sensors in a plug and play manner. This feature is mandatory for easy passage from one application to another or from one site to another. In a dynamic reconfiguration phase the platform must adapt itself in real-time. This feature is mandatory for robustness purposes when working 7 days a week 24 hours a day.

 

2.2 Interactivity

The intelligent cognitive vision platform will be able to adapt its behaviour directly by taking into account end-user specifications. In particular a high level language based on an ontology will allow to describe new classes of objects, new events or new complex activities.

 

2.3 Communication

The intelligent cognitive vision platform must have scalable visualization tools ranging from very simple palm pilot display  to immersive environments, real time transmission of pertinent data and  complex image/video annotation for archiving.

 

3 How can we reach this objective ?

 

This objective is very ambitious, however current state of the art techniques in cognitive vision have shown that they can  propose partial solutions. There is a need to gather the best techniques in a unique framework and to define standards. The following research activities must be performed to reach this objective:

 

3.1 To build a standard state of the art vision library

 

3.2 Program supervision techniques

 

3.3 Real time spatio-temporal reasoning

 

3.4 Ontology of visual concepts

 

3.5 High level language for end-user specification

 

3.6 Learning

 

3.7 Simulation techniques

 

4 Target applications

 

This platform will be a solution for several challenging applications. Among them we can cite smart environments, visual surveillance and virtual teacher.

 

5 Conclusion

 

This ambitious objective is by definition a way to federate cognitive vision research and to concretize the results in a common platform to show that cognitive vision works effectively and allow to start a new generation of vision systems.