Unstructured road detection using hybrid features

Road detection is a key step of the autonomous guided vehicle system such as road following. In this paper, a novel unstructured road detection method is proposed. First, white balance and gray level stretch technique are adopted to enhance image performance. Then, a small overlapped sliding window is scanned over the frame from which hybrid features are extracted. Next, a SVM-based classifier is employed to distinguish the road area from background. At last, the morphological operation and moving average filter technology are performed to obtain precise location of the road region. The proposed algorithm has been evaluated by different type of unstructured roads and the experimental results show its effectiveness.