00100			VISUAL PERCEPTION AND AUTOMATION
00200	
00300	
00400	Minsky and Papert argue that vision is a prime problem area for
00500	the study of intelligence.  They see the main problems as the description
00600	of data and procedures to represent a variety of kinds of knowledge.  This
00700	knowledge we now conceive of in terms of microtheories, well worked out
00800	systems of knowledge in small domains such as topology, geometry, chemistry,
00900	biology, psychology and human relations.  We come to see the problem as
01000	implementing these microtheories as the semantics of particular domains.
01100	The most important microtheories of vision are related to representations.
01200	Two representations help us with problems like:
01300		how do we describe a complex object
01400		how do we relate a scene just after moving our head to that
01500			just before.
01600	These are the representations of shape and space.  As problem areas,
01700	the microtheories of vision are relatively well-defined, since they
01800	are derived from models of topology and geometry, and properties of the
01900	physical world.  Some of the complication of social and psychological models
02000	does not widely enter in vision.  
02100	
02200	Some areas which have been attacked have been:
02300		theoretical understanding of line-drawings  Huffman,Clowes,Waltz
02400		structural learning	Winston
02500		use of world model for outdoor scenes	Yakimovsky
02600		representation of complex objects	Binford,Agin,Nevatia
02700		computational geometry-computational complexity
02800	There is a great need and opportunity for work in perceptual strategies
02900	which has the advantage of a well-defined problem domain for which we
03000	have great intuition, so that we can use semantics to do otherwise very
03100	hard problems.  There is also a need for good systems work.
03200	
03300	This is an applications area with promise.  We have a serious effort in
03400	programmable automation.  The application area is a source of good problems
03500	in AI at this stage.  We hope that more support for AI will come from the
03600	realization of the impact of these techniques on problems outside the
03700	university.
     

00100	using our representation of complex objects, describe a few objects for
00200		example, pump base
00300	
00400	extend the ideas of Huffman and Clowes with dual graphs (see Mackworth))
00500	
00600	fourier transform edge verifier
00700	
00800	Texture analysis:
00900	
01000		2-d shape description by projection
01100		cluster descriptors
01200	
01300	improve convergence of calibration model by choosing improved non-singular
01400		parameters
01500	
01600	stereo correlation, analysis and programming
01700	
01800	edge finder for curves
01900	
02000	use visual feedback to put a screw into a hole.