Several kinds of internal representation have been employed as the output from machine vision systems, depending to some extent on the task to be performed. Similarly, there have been many proposals for the kind of internal representations that may be generated by the human visual system for the wide range of tasks for which it is used.
Chapter 3 introduced the idea of representing the world by a data base of facts. For the purpose of describing the location of the Underground sign to the transport sub-system of the MTG, a fact like the following may be all that is necessary:
[SIGN at [35 40 10] ]
The three numbers specify the position of the sign with respect to the MTG in terms of the distance to the sign, the angle between the direction in which the MTG is facing and the direction from the MTG to the sign, and the height of the sign above ground.
A general purpose vision system will need many such descriptions, and much more besides, in order to navigate a robot around a cluttered environment containing many objects. For the remainder of this chapter, we examine two methods for producing a part of this rudimentary kind of representation, namely, the fact a particular object is visible to the TV camera. In particular, we shall be concerned with the part of the MTG's visual sub-system for recognizing Underground signs. The appendix to this chapter shows how to implement the first method and part of the second method in POP-11. We concentrate on the problem of recognizing the Underground sign in a single image and do not address the problems and benefits of recognizing objects through sequences of TV images.