Motivation

 The projection of light rays onto the retina presents our visual system with an image of the world that is inherently two-dimensional, yet we are able to interact with the three-dimensional world, even in situations new to us, or with objects unknown to us. That we accomplish this task easily implies that one of the functions of the human visual system is to reconstruct a 3-D representation of the world from its 2-D projection onto our eyes.

Ever since the introduction by Julesz of the random dot stereogram as a research tool, it has become commonplace to consider the theoretical problems surrounding stereopsis within a conceptual framework that has two distinctive characteristics:

• First, it is recognized that disparity information can be extracted using only low-level monocular `point' descriptions as the entities which are binocularly matched. High-level monocular descriptions, such as those dealing with surfaces and objects, are not a necessary requirement for stereopsis because for a random dot stereogram these relatively high-level scene descriptions appear only after stereopsis has been achieved.
• Second, it has become acknowledged that there is a need for mechanisms capable of selecting the correct binocular point-for-point matches. That is, in the terminology of Julesz, it is necessary to posit `global stereopsis' mechanisms to resolve ambiguity often existing between competing `local stereopsis' matches.

In this lecture, we will present all the possible problems that one may encounter in shape from stereo vision and research that have been reported to overcome these problems.