Domain-dependent reasoning for visual navigation of roadways

Abstract

A visual navigation system for autonomous land vehicles has been designed at the Computer Vision Laboratory of the University of Maryland. This system includes several modules, among them a 'knowledge-based reasoning module' that is described. This module utilizes domain-dependent knowledge (in this case, 'road knowledge') to analyze and label the visual features extracted from the imagery by the image processing module. Knowledge and general hypotheses are given. The reasoning module itself is described and results are presented. Some conclusions and future extensions are proposed. >