Checking now the automation profile in Table | we may raise the question 
of whether we can do any better or whether the level reached reflects the 
maximum that we reasonably may expect. If we define the end result of 
the mapping project at hand as a symbolic, geometrically correct 
representation of certain objects in the real world, as specified in the 
map contract, then we have the same objective as computer vision: “vision 
Is the process of discovering from images what is present in the world, 
and where it is" (Marr(82)). Before we jump to conclusions and proclaim 
computer vision as the ultimate method to be used in photogrammetry we 
shall have a closer look at the problem. 
The Map Machine 
The hypothetical map machine, depicted in Figure 1, is capable of carrying 
out the map project entirely automatically. It reads the map contract, 
determines all necessary parameters for a sensing device, receives data, 
either on-line or off-line, analyzes and preprocesses the information 
retrieved, and finally generates a digital map according to contract 
specifications and map standards as they apply to the project. 
The map machine should not be taken literally; it is intended to serve as a 
concept or even as a theory about making maps from images. This may 
seem unnecessary since we have known how to make maps for decades; 
hence we may argue the theory exists or is not needed. 
For some of the tasks we have indeed a sound theory. Aerotriangulation, 
for example, is not only well understood, but detailed knowledge ís 
available to instruct a machine to solve it. On the other hand, there are 
problems we solve with great ease, but we struggle badly when it comes 
to instructing a computer to perform them. If we instruct a draftsman to 
generalíze al! natural lines, then he does it without much conscious effort, 
including all the small adjustments needed to preserve consistency. If he 
detects a building in a street intersection he quickly remedies the 
mistake. It is easy to generate an impressive list of problems which 
humans solve easily but which seem to defy computer solutions. Hence, the 
theory about map making appears to be incomplete; otherwise we would be 
in a position to build a machine to do it. 
At first glance one may suspect that we fail to solve certain problems by 
computers only because we do not apply the right methods. While this may 
be a valid observation (we are so entrenched in the algorithmic world of 
statistics and analytical geometry, for example), | do not think it is 
473