Road Boundaries Extraction Using a Centre-Surround Cell Operator and Phase Congurency

In this paper, we present a novel system for extracting road map from pyramid data sets generated from high resolution images. With the model of two Gaussian functions, we design a centre-surround cell operator and subsequent half-wave rectification may help us to find all the pixels on road surfaces. Meanwhile, some dot patterns and step-shaped structures are also incorrectly detected. These misclassified targets are removed by an improved version of line segment matching scheme. The detected road pixels consist of broken road bands are connected. At last, road edge lines are extracted from road bands by making use of phase congruency, this way may restrain the parts inflated before. Experiments demonstrate that the proposed method can work efficiently.