Unstructured Lane Identification Based on Hough Transform and Improved Region Growing

Abstract

With the continuous development of intelligent driving technology, vision-based navigation technology is becoming increasingly significant. Lane detection is the key part of the navigation technology. At present, most of the researches focus on the identification of structured lane, such as highways. However, study on the detection of unstructured lane is relatively few, mainly because unstructured lanes have no obvious marks, which brings difficulties in identification. Therefore, unstructured lane detection still faces serious challenges now. In this paper, region of interest (ROI) of original RGB image is first selected. Next step is RGB image conversion to a grayscale. Then, Gaussian filter and mean filter are used to smooth the structured lane image and unstructured lane image, respectively. Finally, applying edge detection and Hough transform (HT) to extract the unstructured straight lane. Meanwhile, improved region growing and least square (LS) are used to extract unstructured curved lane. The experiment results on the roads of a campus show that the presented algorithms are effective, which can accurately extract the road area and road boundary.