Lane Detection of Unstructured Roads Based on WS-P2PNet

Lane detection, as a foundational computer vision task, is the key technique in autonomous lane keeping system. Compared with structured lane which has clear boundary information, Unstructured lanes such as rural road and urban lane are hard to detect since they lack distinct recognized symbols. This study proposes a new network to detect this kind of unstructured lanes. First, the improved pix2pix named WS-P2PNet is used as backbone network. It can predict the location of the lane. Then, then bottom layer of WS-P2PNet is followed by SCNN which is used to enable message passing between pixels across rows and columns in a layer. Finally, an extensive evaluation of the proposed method is conducted over CULane, and compared with different kind of lane detection algorithms. The results demonstrate that the proposed method achieves a highly satisfactory performance.