The direction of research at CVL has been on methods for development of the complex models required for a cognitive system to understand a complex world. It is a firm belief that systems have to be organized in a hierarchical or pyramid manner, where lower level operation outputs feed into higher level operations. Similarly, higher level information will be used for control of lower level operations in the form of context. Learning has to be used to overcome the immense complexity in specification of a multiplicity of partly overlapping models in dependence upon context. This generates particular demands upon the information representation, to allow information to be communicated and usable at levels different from where it was generated.

It is argued that information must be acquired actively by the system itself, through response driven association with percept transformations. After an active training, however, the system can exhibit a reactive behavior to passively observed percepts. From this derives our view that the development of powerful Cognitive Vision structures inevitably has to go the path over Active Vision, Learning and Robotics, even if the systems will be used for interpretation of static imagery.