Automated Stereo Perception

摘要

This research aims to advance the science of computer stereo vision - reconstruction of 3-dimensional scenes from 2-dimensional images. The problem is to establish a correspondence between features or regions in two or more images from which we can calculate positions in 3-space. In his research, the author chooses to match features, rather than use the traditional methods of area correlation. The features he uses are extended edges, or more precisely, the image curves which are the projections of edges in the scene. Assumed is a preprocessing stage which can extract such edges by first applying an edge operator and then linking edge elements into extended image curves. The choice of feature-based stereo over area-based stereo offers advantages in speed and accuracy, as well as avoiding some fundamental problems.