Visual routines for outdoor navigation

This paper describes several visual routines which can be used to assist the navigation of a car in an outdoor environment, such as streets and highways. These routines are based on the grouping of edge chains in two classes of straight segments and on a suitable segmentation of the image in different regions. Such routines are able to provide a preliminary recovery of the 3D structure of the viewed scene, in which the sides of the road and the buildings are detected, and to recognize several landmarks, such as street signs, trees and light lamps. These results suggest that it is possible to construct drivers' assistant systems based on computer vision technologies.