1 Introduction

This paper presents a cognitive model of a vision system based upon a view of control theory that has been proposed as a rationale for perception and behaviour in living systems. The underlying principle of the theory is that the function of behaviour (output) is to achieve and regulate (control) a certain perception (input). The structure that this scheme imposes on any model is quite different from the more conventional design of computing a specific output from an input. The benefits and implications of this approach are discussed below.

In the present context control theory has been applied to model visual behaviour of object-based fixation. Given a scene containing a number of objects, the purpose of the model, using a robot/camera system, is to fixate a known object from a viewpoint consistent with that of the model object. The main requirements for such a system are,

• a learning scheme that extracts the set of features specific to a model object
• a representation of feature vectors that encodes a scene or object
• a mechanism that measures the correlation between the representations of the current view and that of the model
• a controller that produces behaviour which minimises the mismatch between the object model and the current scene.
• a hierarchy of feature levels analogous to increasing levels of complexity in the visual cortex, providing added object discrimination.

The feature investigated in this paper is a fixation measure which allows control of a vision system active in three dimensions.