A robust lane detection method for autonomous car-like robot

Abstract

Due to illumination variation, view changes, and complex road conditions, automatic lane detection is crucial for path finding and planning for autonomous car-like robots. In this paper, a robust lane detection method is proposed. Firstly, in order to extract the edges of lanes in traffic scenarios, we adopt an adaptive thresholding strategy to binarize a gradient image and trace edges by using their local gradient information. Moreover, by integrating gradient constraints and introducing a reverse voting strategy to the standard Hough transform, we greatly improved speed and stability of line extraction. Further, through inverse perspective mapping the endpoints of extracted lines to the world coordination, we can combine the extracted lines from different cameras. Finally, the lane could be detected by matching two points instead of two parallel lines in parameter space. Extensive experiments and comparisons show the efficiency of the proposed method.