This paper describes a system that gives a simplified interpretation o
f real scenes through analysis of monocular images. Scenes are describ
ed by fusing information derived by the edge segments and by the regio
ns extracted from the scene. The interpretation is given in terms of p
rimitives such as surfaces and their spatial relationships. Real indoo
r and outdoor scenes relative to urban and rural environments have bee
n used to validate the system. The main goal of the proposed architect
ure is to provide help for blind people's navigation, but the techniqu
es developed are also suitable for robotics applications. (C) 1997 Els
evier Science B.V.