Rough Set based Unstructured Road Detection through Feature Learning

Abstract

This paper addresses the navigation problem of robot vehicle, Patrol-security Robot, on unstructured roads with degraded surfaces and edges, strong shadows and no lane markings. These conditions cause many road-following systems failed because the road feature extraction is not easy and the detection effect becomes inaccuracy. For solving the problem of the feature extraction of the complex unstructured road detection, a rough set based unstructured road detection (RSURD) method is presented. By learning from the samples of road images, the knowledge of road features is acquired and consummated gradually by on-line learning, and the accuracy of rules is improved, which make the system fit different road conditions. This method has been implemented and tested on Patrol-security Robot. Good adaptability, robustness and reliability have been accomplished.