Abstract

This paper presents the current machine vision paradigm as an attempt to solve five basic problems concerned with different aspects of modeling the physical environment from 2-D images.

These problems are:

(1) recovering scene geometry,

(2) image partitioning,

(3) object labeling,

(4) relational description in both space and time, and

(5) system architecture.

We describe the reason for the difficulty of each of these problems, and provide examples of novel approaches currently being investigated as potential solutions.

© 1986 Optical Society of America